Variable Rate Image Compression based Adaptive Data Transfer

Algorithm for Underwater Wireless Sensor Networks

Bin Wang and Kerong Ben

College of Electronic Engineering, Naval University of Engineering, Wuhan, China

Keywords: Underwater Wireless Sensor Networks, Underwater Image, Routing Algorithm, Reinforcement Learning

(RL), Haar Wavelet Transform Algorithm.

Abstract: Underwater image, as a kind of important data in underwater wireless sensor networks, can more

comprehensively and intuitively reflect the state of underwater environment, but the traffic demand of

which is greater than the demand of the numerical data. However, the underwater acoustic channel has the

characteristics of high bit error rate, high delay, low bandwidth and so on, and the energy of communication

nodes is limited, and the position is time-varying, which makes the transmission of underwater image data

extremely difficult. Aiming at this problem, this paper proposes an adaptive image transfer algorithm for

underwater wireless sensor networks. The algorithm is based on HAAR wavelet transform algorithm to

provide multi-resolution image lossless compression, and can adapt to different image transmission bit rates

according to the changes of underwater transmission conditions. It can intelligently select the transmission

routes based on reinforcement learning algorithm to achieve reliable and efficient underwater image transfer.

Experiments show that the algorithm can effectively improve the packet delivery rate of underwater image,

reduce the transmission delay and energy consumption, and can distinguish the transmission of image

feature data and detail data, and balance the distribution of communication energy consumption among

underwater communication nodes.

INTRODUCTION

Underwater wireless sensor networks are generally

composed of data centers, water surface sink nodes,

underwater communication nodes and underwater

sensor nodes (Qiu, 2020), which rely on the

self-organizing communication ability between

nodes to achieve the data transfer. Underwater

wireless sensor networks are widely used in many

fields such as accident warning, ecological

monitoring, hydrological data collection, marine

resource exploration, auxiliary navigation and so on

(Qiu T, Jiang S.). At present, underwater

long-distance wireless communication is still

dominated by underwater sound. Compared with the

terrestrial wireless sensor network based on the

radio, the underwater transmission channel has the

characteristics of small capacity, large delay, more

noise and interference factors, and the

communication node has the characteristics of

limited energy, difficult supply, and high

spatiotemporal dynamics (Fattah and Haque), which

makes it extremely difficult to achieve the reliable

underwater communication.

Underwater images can truly, accurately, and

continuously reflect the real-time situation of the

monitoring area. They are widely used in the

monitoring of marine perimeter, important

underwater facilities, and marine biodiversity. They

are also an important basis for underwater target

positioning and recognition (Gupta and Boukerche).

The quality of underwater images and the

transmission efficiency from underwater to water

surface greatly affect the effect of follow-up

monitoring and research work. Compared with the

numerical data, the underwater image data is larger,

and the underwater image data transfer is more

difficult. However, there is some redundancy in the

underwater image data. The image compression

algorithm based on HAAR wavelet transform can

realize the lossless compression of the underwater

image data. And it can not only reduce the data

transfer demand, but also adjust the transmission

data type and the transmission code rate according to

the change of transmission conditions and optimize

the output underwater image quality (Menon and

Kanagaraj).

Wang, B. and Ben, K.

Variable Rate Image Compression Based Adaptive Data Transfer Algor ithm for Underwater Wireless Sensor Networks.

DOI: 10.5220/0012273900003807

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 2nd International Seminar on Artiﬁcial Intelligence, Networking and Information Technology (ANIT 2023), pages 61-69

ISBN: 978-989-758-677-4

Reinforcement learning (RL) refers to the

process in which agents learn a mapping relationship

from environmental state to behavior by

continuously interacting with the environment and

using the reward of environmental feedback [Guo

W, Chen Y]. Nodes in underwater wireless sensor

networks can update the status of the transmission

environment based on the information interaction

between neighbor nodes, and then determine the

current reward and cumulative reward after each

action. By calculating the status value obtained, the

strategy for performing the next action is determined

(Zhou Y-Su Y). The transmission route of

underwater network needs to be dynamically

selected according to the change of transmission

conditions. After reinforcement learning algorithm is

applied to underwater wireless sensor networks, it

can provide decision-making basis for underwater

image data when selecting the next hop node

according to underwater transmission conditions and

historical data forwarding.

For the underwater target recognition

applications, this paper presents an adaptive data

transfer algorithm for underwater wireless sensor

networks based on variable bit rate image

compression and reinforcement learning

(VRC-ADTA). In this algorithm, HAAR wavelet

transform is used as variable rate image compression

algorithm, and Q-learning is used as reinforcement

learning algorithm. The reward function is designed

based on the position of underwater communication

nodes, residual energy, and transmission delay. The

variable transmission code rate adjustment and

dynamic routing of underwater image data are

realized. The simulations show that VRC-ADTA can

satisfy the accuracy requirements of underwater

target recognition, improve the efficiency of

underwater image data transfer, reduce the

transmission delay, the energy consumption, and

improve the packet delivery rate.

RELATED WORKS

To overcome the unfavorable conditions of

underwater network, researchers have studied and

implemented efficient and reliable data transfer

routing algorithms for underwater wireless sensor

networks from different perspectives. Table 1 lists

several typical routing algorithms and their

characteristics, among which communication

efficiency mainly requires routing algorithms to

control transmission delay and reduce

communication energy consumption.

Table 1: Several typical data transfer algorithms for

underwater wireless sensor networks.

Algorithm

Establishment

Metho

pplicati

Efficien

VBF

[Xie P]

Routing based on node

location.

istinctio

Low

QELAR

[Hu T]

Routing based on the

node location and

residual ener

istinctio

Low

RCAR

[Jin Z]

Routing based on the

delay and residual

ener

Business

flow

Middle

EP-ADTA

[Wang B]

Routing based on the

node location, the

delay, and the residual

ener

Content

prediction

High

RC-AD

Routing based on the

node location, the

delay, and the residual

ener

djustabl

code rate

High

The VBF proposed by Xie et al. belongs to the

geographical routing protocol. According to the

position vector between the sending node and the

target node, the node is selected as the next hop, but

it cannot balance the energy consumption between

the neighbor nodes, making the nodes closer to the

position vector consume more energy due to

excessive forwarding. QELAR proposed by Hu et al.

builds the transmission routes based on

reinforcement learning, and adaptively selects the

best route to reduce the transmission hops and

balance the energy consumption. Both VBF and

QELAR can find a suitable underwater route to

realize data transfer without distinguishing between

the carried applications. However, it is necessary to

adjust the transmission route reasonably according

to the priority of applications and the queue length

of nodes.

The RCAR proposed by Jin et al. focuses more

on the congestion avoidance method in the case of

large traffic. Through the reinforcement learning

algorithm, it optimizes the distribution of the delay

and energy consumption on the transmission route,

and can well adjust the traffic distribution in case of

heavy traffic. Although RCAR optimizes the traffic

layout in time-varying underwater networks and can

provide a good QoS, it is not enough to only

consider improving the unfavorable underwater

transmission conditions. It is also necessary to

consider how to better integrate the business

applications to improve the efficiency of underwater

communication.

ANIT 2023 - The International Seminar on Artiﬁcial Intelligence, Networking and Information Technology

The EP-ADTA proposed by Wang et al. is

designed for the transmission of the time series

monitoring data. While selecting the transmission

route, it adaptively adjusts the transmission accuracy

of the monitoring data according to the transmission

conditions. When the underwater environment is

good and the transmission route can provide enough

transmission bandwidth, high-precision monitoring

data can be transmitted. When the transmission

environment is bad and the transmission route

cannot provide enough bandwidth, the transmission

accuracy of monitoring data shall be appropriately

reduced to ensure that the characteristic data is

transmitted to the surface sink node in priority.

Therefore, EP-ADTA considers the reliability and

efficiency of transmission. However, EP-ADTA can

only be applied to the time series data acquisition,

and is not applicable to the underwater image data.

Therefore, based on the above research, aiming

at the fixed 3D underwater wireless sensor network,

in order to improve the efficiency of underwater

image data transfer, the research of adaptive data

transfer algorithm based on variable bit rate image

compression technology (VRC-ADTA) is carried

out.

ALGORITHM DESIGN

VRC-ADTA consists of two parts. One is the image

compression algorithm based on the HAAR wavelet

transform, which realizes lossless compression of

image data and converts the image data into the

average feature data and the detail coefficient data.

The image compression algorithm is implemented in

underwater sensor nodes. The second is the routing

algorithm based on the reinforcement learning,

which realizes the optimal underwater transmission

route selection. The routing algorithm is

implemented between the underwater sensor node

and the communication node.

3.1 Image Compression Algorithm Based on

HAAR Wavelet Transform

The image, especially the static continuous tone

image, has great redundancy between adjacent

pixels. The change of image sample values between

adjacent pixels is smooth and generally does not

change suddenly, even if there is a mutation, it is

only at the edge of the object in the image, which

makes it possible to compress images. In order to

compress the image effectively, we must focus on

the redundancy contained in the signal and reduce its

redundancy. In the usual image, most of the signals

are concentrated in the lowest frequency component.

The higher the frequency, the more the signal

strength decays.

Due to the scalable resolution of the discrete

wavelet transform, the image compression algorithm

based on wavelet transform can transform the

resolution repeatedly in space, and reduce the

redundancy of the image and compress the image.

HAAR wavelet transform is a typical wavelet

transform, which takes the average value of two

adjacent values as the low-frequency coefficient and

the difference between two adjacent values as the

high-frequency coefficient [Xiang W, 20]. When the

adjacent values are equal or the change rate is very

gentle, the low-frequency coefficient can be used as

the approximation of the sample value, while the

high-frequency coefficient is a value close to 0. In

this way, in some cases, the high-frequency

coefficient can be discarded and the low-frequency

coefficient can be directly used to restore the image.

The restored image cannot reach the quality of the

original image. In many cases, this loss of image

quality can fully meet the needs of practical

applications (Bagmanov V H, Porwik P).

Suppose a one-dimensional image with a

resolution of only 4 pixels, and the corresponding

pixel values are respectively: [9 7 3 5], calculate its

HAAR wavelet transform coefficients.

Step 1: calculate the average value. Calculate the

average value of adjacent pixel pairs to get a new

image with relatively low resolution, whose number

of pixels has become 2. That is, the resolution of the

new image is 1/2 of the original, and the

corresponding pixel value is [8 4], where 8= (9+7)/2,

4= (3+5)/2.

Step 2: calculate the difference value. When this

image is represented by 2 pixels, the image

information has been partially lost. To reconstruct

the original image composed of 4 pixels from the

image composed of 2 pixels, it is necessary to store

the image detail coefficients for retrieving the

missing information during reconstruction. The

method is to subtract the average value of the pixel

pair from the first pixel value of the pixel pair. The

first detail coefficient is (9-8) =1, because the

calculated average value is 8, which is 1 smaller

than 9 and 1 larger than 7. Storing this detail

coefficient can restore the first two-pixel values of

the original image. Using the same method, the

second detail coefficient is (3-4) =-1, and the last

two-pixel values can be restored by storing this

detail coefficient. Therefore, the original image can

Variable Rate Image Compression Based Adaptive Data Transfer Algorithm for Underwater Wireless Sensor Networks

be expressed as [8 4 1 -1] with the following two

average values and two detail coefficients.

Step 3: repeat steps 1 and 2 to further decompose

the image obtained from the step 1 into the images

with lower resolution and the detail coefficients. At

last, the whole image is represented by the average

value of one pixel 6 and three detail coefficients 2, 1

and -1. The decomposition process is shown in

Table 2.

Table 2: Schematic diagram of HAAR wavelet transform.

Resolving

powe

Average

value

Detail

facto

4 [9 7 3 5]

2 [8 4] [1 -1]

[6] [2]

It can be seen from Table 2 that the image

composed of four pixels is represented by one

average pixel value, one first-order detail coefficient

and two second-order detail coefficients through the

above decomposition.

After the underwater image data is compressed

based on HAAR wavelet transform, the pixel data of

the image is transformed into a combination of the

average feature value and the detail coefficient

values. And the data combination of the average

value and the detail coefficient values of the

transmitted image can restore the image of the

previous level of resolution. Only transmitting the

average value can retain the feature of the image and

reduce the demand for data transfer. After the detail

coefficient data arrives at the later stage, the

underwater image quality can be further improved.

3.2 Routing Algorithm Based on

Reinforcement Learning

Reinforcement learning is a machine learning

algorithm aimed at finding the optimal mapping

strategy from state to action, typically used to solve

problems related to Markov decision processes

(MDP) (Dugaev, 2020). The process of

reinforcement learning is usually represented by five

tuples



𝑆, 𝐴, 𝑃, 𝑅, γ



(Jin Z, Shen Z), where S is the

environment, A is the action, P is the transition

probability, R is the reward, and γ is the discount

rate.

Assuming that the underwater wireless sensor

network is composed of m nodes, the nodes can be

expressed as:



𝑛



, 𝑛



, ⋯, 𝑛





(1)

Where, 𝑛



represents underwater wireless

sensor network nodes, and m represents the number

of nodes. Each the node of the underwater network

can obtain its own coordinate. Then, the candidate

relay node set of the current node 𝑛



can be

expressed as:

𝑁





𝑖





𝑛



𝑁|𝑑𝑛



−𝑑



𝑛





≤0

∩𝑛𝑒𝑖



𝑛







(2)

Where, 𝑁



(𝑖) is the set of candidate relay

nodes, 𝑛𝑒𝑖

(

𝑛



)

is the neighbor nodes set covered

by one hop, 𝑁|𝑑𝑛



−𝑑

(

𝑛



)

≤0 represents the

node set which is shallower than the depth of the

current node.

If the packet is located at the node 𝑛



, the

current environment state S can be defined as:







∪N



(

)

(3)

The action A can be defined as:





∈S



(4)

If the current packet is at the node 𝑛



and 𝑛



selected as the relay node, the reward function is:

𝑅













= −𝑅



−



𝜑



× 𝑐𝑜

(

𝑒

)

+ 𝜑



× 𝑐𝑜

(

𝑡

)

(5)

The reward function R includes fixed cost,

residual energy cost of neighbor node and

transmission channel delay cost in three parts.

The significance of 𝑅

is that the fixed cost

needs to be increased every time the data forwarding

of a hop node is experienced. The existence of 𝑅

can help the agent to select the route with fewer

hops.

The significance of 𝑐𝑜

(

𝑒

)

is that selecting the

node with large residual energy as the next hop node

is beneficial for extending the service time of the

underwater networks. The existence of 𝑐𝑜

(

𝑒

)

can

help the agent select the node with greater residual

energy as the next-hop node. It can be expressed as:

(

𝑒

)

=1−

𝐸





∑

𝐸





∈



(



)

(6)

Where, 𝐸





represents the residual energy of the

next-hop node, and

∑

𝐸





∈



(



)

represents the total

residual energy of the candidate node set, 𝜑



is the

sensitivity coefficient for the 𝑐𝑜

(

𝑒

)

The significance of 𝑐𝑜

(

𝑡

)

is to select the

transmission channel with smaller transmission

delay to reach the next-hop node, and the smaller

transmission delay shows that the transmission

channel is more stable, reliable, has less bit error rate

and congestion. It can be expressed as:

ANIT 2023 - The International Seminar on Artiﬁcial Intelligence, Networking and Information Technology

𝑐𝑜

(

)

=1−

𝑡



→

(7)

Where, 𝜑



is the sensitivity coefficient of the

𝑐𝑜

(

𝑡

)

, 𝑡



→

is the total transmission delay of the

packet.

If the current packet is at the node 𝑛



, the

transition probability of the node 𝑛



to the node

𝑛



is defined as:

























∑

















∈

(8)

In order to embody the impact of the next state

on the current state, the overall reward is defined as:

𝑅



= 𝑟



+ 𝛾𝑟



+ 𝛾



𝑟



+ ⋯

= 𝛾



𝑟







(9)

According to the Q-learning, the state action

function under the policy π is defined as:

𝑄



(

𝑠, 𝑎

)

= 𝐸





𝑅



|𝑠



= 𝑠, 𝑎



= 𝑎



(10)

Assuming that sum is the next action a

′

and the

next state s

′

, the optimal solution 𝑄

∗

(

𝑠, 𝑎

)

in the

state 𝑄



(

𝑠, 𝑎

)

can be expressed as the iterative

equation:

𝑄

∗

(

𝑠, 𝑎

)

= 𝑟



+ 𝛾 𝑃









max





𝑄

∗

(

𝑠



, 𝑎



)







∈

(11)

The V value function will select the Q value that can

obtain the maximum benefit, which is defined as:



∗

(

𝑠

)

=max



𝑄

∗

(

𝑠, 𝑎

)

(12)

At the initial stage, the Q value table of each

agent is initialized according to its neighbor

relationship and the position to improve the

convergence speed of the Q value. The initial setting

of Q value is:

𝑄





→







= −10 , 𝑛



, 𝑛



∈𝑁, 𝑛



∉𝑁



(

𝑖

)

(13)

As the Q values and V values are gradually

updated and converged according to formulas (11)

and (12), a good data transmission strategy will

eventually emerge.

The reinforcement learning adjusts the degree of

"exploration" and "utilization" by exploring

probability 𝜀 to ensure that the best strategy can be

used without missing the global

best strategy

(El-Banna A A A, 2021). When the exploration

probability 𝜀 is small, the reinforcement learning

algorithm will choose more random strategies to

explore new transmission route. When the

exploration probability 𝜀 is large, the reinforcement

learning will use more existing optimal strategies to

achieve more efficient and reliable data transfer.

When transmitting the average value data, because it

contains the core features of the image, we choose a

larger probability of exploration 𝜀, use the existing

optimal strategy for transmission, and increase the

data retransmission threshold 𝑡𝑖𝑚𝑒𝑠

ℎ

improve the reliability of the image feature data

transfer. When transmitting the detailed data, choose

a smaller exploration probability 𝜀, and try more

random strategies to reduce the energy consumption

of the optimal path. At the same time, try to obtain

the global optimal path and reduce the data

retransmission threshold 𝑡𝑖𝑚𝑒𝑠

ℎ

. The

corresponding relationship between the explore

probability 𝜀 , the retransmission threshold

𝑡𝑖𝑚𝑒𝑠

ℎ

and the image data type, which is

defined as:

𝜀

= 

𝜀



𝑤ℎ𝑒𝑛 𝑡ℎ𝑒 average 𝑣𝑎𝑙𝑢𝑒𝑠 𝑎𝑟𝑒 𝑓𝑜𝑟𝑤𝑎𝑟𝑑𝑒𝑑

𝜀



, 𝑤ℎ𝑒𝑛 𝑡ℎ𝑒 𝑓𝑖𝑟𝑠𝑡 𝑑𝑒𝑡𝑎𝑖𝑙 𝑓𝑎𝑐𝑡𝑜𝑟𝑠 𝑎𝑟𝑒 𝑓𝑜𝑟𝑤𝑎𝑟𝑑𝑒𝑑

𝜀



, 𝑤ℎ𝑒𝑛 𝑡ℎ𝑒 𝑠𝑒𝑐𝑜𝑛𝑑 𝑑𝑒𝑡𝑎𝑖𝑙 𝑓𝑎𝑐𝑡𝑜𝑟𝑠 𝑎𝑟𝑒 𝑓𝑜𝑟𝑤𝑎𝑟𝑑𝑒𝑑

(14)

𝑡𝑖𝑚𝑒𝑠





𝑡𝑖𝑚𝑒𝑠



, 𝑤ℎ𝑒𝑛 𝑡ℎ𝑒 average 𝑣𝑎𝑙𝑢𝑒𝑠 𝑎𝑟𝑒 𝑓𝑜𝑟𝑤𝑎𝑟𝑑𝑒𝑑

𝑡𝑖𝑚𝑒𝑠



, 𝑤ℎ𝑒𝑛 𝑡ℎ𝑒 𝑓𝑖𝑟𝑠𝑡 𝑑𝑒𝑡𝑎𝑖𝑙 𝑓𝑎𝑐𝑡𝑜𝑟𝑠 𝑎𝑟𝑒 𝑓𝑜𝑟𝑤𝑎𝑟𝑑𝑒𝑑

𝑡𝑖𝑚𝑒𝑠



, 𝑤ℎ𝑒𝑛 𝑡ℎ𝑒 𝑠𝑒𝑐𝑜𝑛𝑑 𝑑𝑒𝑡𝑎𝑖𝑙 𝑓𝑎𝑐𝑡𝑜𝑟𝑠 𝑎𝑟𝑒 𝑓𝑜𝑟𝑤𝑎𝑟𝑑

𝑒

(15)

3.3 Simulation and Performance

Analysis

The simulation is based on the underwater fish

activity monitoring application. The simulation is set

as a three-dimensional underwater area. One sensor

node is deployed in the underwater area center to

collect image data, and several communication

nodes are randomly deployed in the water to forward

the monitoring data. One sink node is deployed in

the surface center to collect the data.

3.4 Simulation and Parameter Setting

In the simulation, the underwater network topology

is generated randomly. Each underwater node

communicates based on the acoustic channels.

Underwater nodes move randomly around their

original positions under the influence of water flow.

Each pixel of the image is represented by 1 byte, and

each packet encapsulates 50-pixel data for

transmission

. The simulation environment and the

Variable Rate Image Compression Based Adaptive Data Transfer Algorithm for Underwater Wireless Sensor Networks

parameter settings of VRC-ADTA are shown in

Table 3.

Table 3: Simulation and the parameter settings.

Paramete

Val u e

Underwater network

500m×500m×500m

Sound speed

1500 m/s

Sound frequency

10 kHz

Communication distance

150m

Number of nodes

100, 200, 300

Maximum distance of

node movemen

0, 5, 10

Initial energy

30000 J

Transmission powe

10 W

Receiving powe

3 W

Channel interruption

probability

0, 0.01,0.1

Image Original Size

(300,500), single

channel

Resolution ratio of images

(Ori

in:1level:2level)

16:4:1

Packet transmission rate

10,15,20

ackets/minute

Ori

in data packet size 100B

tes

Mac protocol S-FAMA

𝑅



-1

𝜑



, 𝜑



0.8, 0.2

𝜀



, 𝜀



, 𝜀



0.7, 0.8,0.9

times



,times



,times



0,1,3

The environment is built by Python. The basic

state of the environment is that the number of nodes

in the underwater network is 100, the data transfer

rate is 10 packets/min, the transmission channel is

uninterrupted, and the node position is unchanged.

3.5 Image Compression Performance

Analysis

As shown in Figure 1-(a), the original image is taken

by an underwater camera. The main body of the

image is fish swimming in the water, and the

background is underwater reef. As shown in Figure

1-(b), the size of the image (1level) after one

compression based on HAAR wavelet transform is

reduced to 1/2 of the size of the original image, and

the pixel data to be transmitted becomes 1/4 of the

original image, but the fish swimming in the water

can still be clearly identified. As shown in Figure

1-(c), the size of the image (2level) after twice

compression based on HAAR wavelet transform is

reduced to 1/4 of the size of the original image. The

pixel data to be transmitted becomes 1/16 of the

original image, and the fish swimming in the water

can still be recognized. Therefore, image

compression based on HAAR wavelet transform

retains the characteristic information of the image.

When the transmission channel is bad and a large

amount of data is not allowed to upload, the image

average data generated by HAAR wavelet transform

retains the main characteristics of the image and can

meet the basic needs of image monitoring. Because

the image compression based on HAAR wavelet

transform belongs to lossless compression, after the

transmission channel is restored, continue to

transmit the image detail coefficient data, which can

completely restore the resolution of the original

image.

Figure 1: Comparison of image compression effects with

different resolutions.

Where, Figure 1-(a) is the original underwater

image; Figure 1-(b) shows the image after one

compression based on HAAR wavelet transform;

Figure 1-(c) shows the image after twice

compression based on HAAR wavelet transform.

3.6 Data Transfer Performance

Analysis

VRC-ADTA, QELAR and VBF are used as routing

algorithms to forward the image packet in

underwater wireless sensor networks, and their

transmission performance is compared.

1) Comparison of the packet delivery rate

According to the different conditions, in underwater

wireless sensor networks, VRC-ADTA, QELAR and

VBF are used to transmit a group of image data

(including 1500 packets), and the average packet

delivery rate of the image packets transmitted from

the underwater sensor nodes to the surface sink

nodes is calculated. The comparative results are

shown in Figure 2.

ANIT 2023 - The International Seminar on Artiﬁcial Intelligence, Networking and Information Technology

Figure 2: Comparison of the average packet delivery rates.

100-Nodes, 150-Nodes, and 200-Nodes respectively

indicate that the network contains 100, 150 or 200

communication nodes; DD-5 and DD-10 indicate

that the dynamic range of node position movement

in the network per minute is 5 meters or 10 meters;

OP-0.01, OP-0.1 indicate that the interruption

probability of each data transfer is 0.01 or 0.1;

TF-15 and TF-10 indicate that 15 or 10 packets are

sent per minute.

Figure 2 shows that under the different

conditions, the blue data column representing

VRC-ADTA is higher than the green data column

representing QELAR and the red data column

representing VBF. This shows that the average

delivery rate of image data in VRC-ADTA is higher

under different conditions. Due to the

comprehensive consideration of transmission hops,

transmission delay, residual energy, and service type,

VRC-ADTA can provide more reliable data transfer

with the support of the adaptive improved

Q-learning.

2) Comparison of the Transmission Delay

VRC-ADTA, QELAR and VBF are used to forward

a group of image data, and the transmission delay of

a single image packet from the sensor node to the

sink node is calculated. The comparative results are

shown in Figure 3.

Figure 3: Comparison of the transmission delay of a single

packet transmitted.

Figure 3 shows that under different conditions,

the blue data column representing VRC-ADTA is

lower than the green data column representing

QELAR and the red data column representing VBF.

This shows that the VRC-ADTA consumes less time

to transmit a single packet under different conditions.

VRC-ADTA can provide faster data transfer. Figure

3 also shows that the increase of interruption

probability has a significant impact on the

underwater data transfer. With the increase of

interruption probability, the data needs to be

retransmitted for many times, resulting in the

increase of the transmission delay.

3) Comparison of the Energy Consumption

VRC-ADTA, QELAR and VBF are used to transmit

a group of image packet, and the communication

consumption required to transmit a single image

packet from the sensor node to the sink node is

calculated. The comparative results are shown in

Figure 4.

Figure 4: Comparison of the energy consumption of a

single packet transmitted

Figure 4 shows that under different conditions,

the blue data column representing VRC-ADTA is

lower than the green data column representing

QELAR and the red data column representing VBF.

It shows that under different conditions, the

VRC-ADTA requires the least energy consumption

for each node of the underwater network to transmit

a single packet. This is mainly because the

VRC-ADTA requires less transmission hops,

indicating that VRC-ADTA can provide more

efficient data transfer. Figure 5 also shows that with

the increase of the number of nodes in the network,

the energy consumption required to transmit a single

image packet increase.

4) Comparison of the Residual Energy Variance

VRC-ADTA, QELAR and VBF are respectively

used to transmit a group of image data, collect the

residual energy of each communication node,

calculate its average and variance of the distribution,

and form a comparison result, as shown in Figure 5.

Variable Rate Image Compression Based Adaptive Data Transfer Algorithm for Underwater Wireless Sensor Networks

Figure 5: Comparison of the average and variance of

residual energy distribution.

Figure 5 shows that VRC-ADTA and QELAR have

a larger average of the residual energy than VBF,

indicating that the two algorithms are more efficient.

The variance of residual energy distribution

corresponding to VRC-ADTA is smaller, indicating

that VRC-ADTA uses more energy evenly than

QELAR and VBF, which is conducive to extending

the overall life of underwater network.

CONCLUSION

The variable rate image compression based adaptive

data transfer algorithm (VRC-ADTA) for

underwater wireless sensor networks proposed in

this paper contains three innovations. Firstly, it is to

introduce the image compression algorithm based on

HAAR wavelet transform into underwater image

transmission. Because HAAR wavelet transform

retains the image features in the process of image

compression, even after the image data is greatly

compressed, the monitoring object is still clear and

recognizable in the image. Because the image

compression based on HAAR wavelet transform

belongs to lossless compression, whether to transmit

the detail coefficient data can be selected according

to the change of transmission channel quality. After

the detail coefficient data is supplemented, the

monitoring image resolution can be restored to the

initial state. Secondly, it is to realize underwater

network routing based on the reinforcement learning.

In the process of the next hop selection, the depth

and residual energy of the relay node, the

transmission channel delay are comprehensively

considered to improve the quality of transmission

routing. Thirdly, it based on whether the content of

the underwater transmission data is the average

value data of the monitoring image or the detail

coefficient data, the exploration probability and the

retransmission threshold in the routing algorithm are

dynamically selected, so that the average value data

containing image features can be delivered to the

destination by the more efficient and reliable route,

while the detail coefficient data that helps to

improve the image resolution are delivered by the

suboptimal routing, in order to reduce the energy

consumption of nodes on the optimal route and

optimize the distribution of the residual energy of

nodes. Therefore, VRC-ADTA can adaptively select

the transmission routes and the resolution of the

image data according to the changes of underwater

transmission conditions. When the transmission

environment is stable, the transmission channel error

rate is low, and the delay is small, it can transmit

high-resolution high-rate image data. When the

transmission environment is unstable and the

transmission channel quality is poor, it can transmit

low-rate image data containing image features, when

feature data and detail data are transmitted at the

same time, it can provide more efficient and reliable

transmission routes for the image feature data.

Simulation shows that VRC-ADTA can provide

efficient and reliable transmission routes under

different node densities, dynamic ranges of node

locations, interruption probabilities of transmission

channels and traffic flows. Compared with QELAR

and VBF, the VRC-ADTA can increase the packet

delivery rate by 3% -20%, reduce the transmission

delay by 10% -50%, reduce the energy consumption

by 18% -60%, and reduce the variance of residual

energy of nodes by 16% -75%.

REFERENCES

Qiu T, Zhao Z, Zhang T, et al. Underwater Internet of

Things in Smart Ocean: System Architecture and Open

Issues[J]. IEEE Transactions on Industrial Informatics,

2020,16(7):4297-4307. https://doi.org/10.1109/TII.201

9.2946618

Jiang S. Networking in Oceans: A Survey[J]. ACM

Computing Surveys, 2021,54(1). https://doi.org/1

0.1145/3428147

Fattah S, Gani A, Ahmedy I, et al. A Survey on

Underwater Wireless Sensor Networks: Requirements,

Taxonomy, Recent Advances, and Open Research

Challenges[J]. SENSORS, 2020,20(18). https://doi.org/

10.3390/s20185393

Luo J, Chen Y, Wu M, et al. A Survey of Routing

Protocols for Underwater Wireless Sensor Networks[J].

IEEE Communications Surveys and Tutorials, 2021,

ANIT 2023 - The International Seminar on Artiﬁcial Intelligence, Networking and Information Technology

23(1):137-160. https://doi.org/10.1109/COMST.2020.

3048190

Haque K F, Kabir K H, Abdelgawad A. Advancement of

Routing Protocols and Applications of Underwater

Wireless Sensor Network (UWSN)-A Survey[J].

Journal of Sensor and Actuator Networks, 2020,

9(2):19. https://doi.org/10.3390/jsan9020019

Gupta O, Goyal N, Anand D, et al. Underwater Networked

Wireless Sensor Data Collection for Computational

Intelligence Techniques: Issues, Challenges, and

Approaches[J]. IEEE ACCESS, 2020,8:122959-122974.

https://doi.org/10.1109/ACCESS.2020.3007502

Boukerche A, Sun P. Design of Algorithms and Protocols

for Underwater Acoustic Wireless Sensor Networks[J].

ACM COMPUTING SURVEYS, 2021,53(6).

https://doi.org/10.1145/3421763

Menon V, Midhunchakkaravarthy D, Verma S, et al.

Enabling Reliable Communication in Internet of

Underwater Things: Applications, Challenges and

Future Directions[C]. 2021 2nd International

Conference on Secure Cyber Computing and

Communications (ICSCCC), 2021, 296–301.

https://doi.org/10.1109/ICSCCC51823.2021.9478081

H. Kanagaraj and V. Muneeswaran. Image compression

using HAAR discrete wavelet transform[C]. 2020 5th

International Conference on Devices, Circuits and

Systems (ICDCS), IEEE. https://doi.org/10.1109/

ICDCS48716.2020.243596

Guo W, Zhang W. A survey on intelligent routing

protocols in wireless sensor networks[J]. Journal of

Network and Computer Applications, 2014,38:185-201.

https://doi.org/10.1016/j.jnca.2013.04.001

Chen Y, Zheng K, Fang X, et al. QMCR: A

Q-Learning-Based Multi-Hop Cooperative Routing

Protocol for Underwater Acoustic Sensor Networks[J].

CHINA COMMUNICATIONS, 2021,18(8):224-236.

Zhou Y, Cao T, Xiang W. Anypath Routing Protocol

Design via Q-Learning for Underwater Sensor

Networks[J]. IEEE Internet of Things Journal,

2021,8(10):8173-8190. https://doi.org/10.1109/JIOT.

2020.3042901

Rodoshi R T, Song Y, Choi W. Reinforcement

Learning-Based Routing Protocol for Underwater

Wireless Sensor Networks: A Comparative Survey[J].

IEEE ACCESS, 2021,9:154578-154599. https://doi.org/

10.1109/ACCESS.2021.3128516

Su Y, Fan R, Fu X, et al. DQELR: An Adaptive Deep

Q-Network-Based Energy- and Latency-Aware

Routing Protocol Design for Underwater Acoustic

Sensor Networks[J]. IEEE ACCESS, 2019,7:9091-9104.

https://doi.org/10.1109/ACCESS.2019.2891590

Xie P, Cui J, Lao L. VBF: Vector-based forwarding

protocol for underwater sensor networks[M]//Boavida

F, Plagemann T, Stiller B, et al. Berlin: Springer-Verlag

Berlin, 2006:1216-1221

Hu T, Fei Y. QELAR: A Machine-Learning-Based

Adaptive Routing Protocol for Energy-Efficient and

Lifetime-Extended Underwater Sensor Networks[J].

IEEE Transactions on Mobile Computing, 2010,

9(6):796-809. https://doi.org/10.1109/TMC. 2010.28

Jin Z, Zhao Q, Su Y. RCAR: A Reinforcement

-Learning-Based Routing Protocol for Congestion

-Avoided Underwater Acoustic Sensor Networks[J].

IEEE Sensors Journal, 2019,19(22): 10881-10891.

https://doi.org/10.1109/JSEN.2019.293 2126

Wang B, Ben K, Lin H, et al. EP-ADTA: Edge

Prediction-Based Adaptive Data Transfer Algorithm for

Underwater Wireless Sensor Networks (UWSNs)[J].

SENSORS, 2022,22(15). https://doi.org/10.3390/s

22155490

Xiang W, Xu W. Study on the wavelet image coding

algorithm [C]. 2011 IEEE 3rd International Conference

on Communication Software and Networks (ICCSN

2011), 2011, 537–541. https://doi.org/10.1109/

ICDCS48716.2020.243596

Batziou E, Ioannidis K, Patras I, et al. Low-Light Image

Enhancement Based on U-Net and HAAR Wavelet

Pooling, 2023[C]. Springer International Publishing

AG, 2023, 510–522. https://doi.org/10.1007/978-3-

031-27818-1_42

Bagmanov V. H., Kharitonov S V, Meshkov I K, et al.

Wavelet multiscale processing of remote sensing data,

2008[C]. SPIE - Optical Technologies for

Telecommunications, 2008, 73740 (8pp.). https://doi.

org/10.1117/12.829035

Porwik P, Lisowska A, Bubak M, et al. The new graphic

description of the HAAR wavelet transform, 2004[C].

Computational Science - ICCS, 2004, 1–8.

https://doi.org/10.1007/978-3-540-25944-2_1.

Dugaev D, Peng Z, Luo Y, et al. Reinforcement-Learning

Based Dynamic Transmission Range Adjustment in

Medium Access Control for Underwater Wireless

Sensor Networks[J]. Electronics, 2020, 9(10): 1727.

https://doi.org/10.3390/electronics9101727

Shen Z, Yin H, Jing L, et al. A Cooperative Routing

Protocol Based on Q-Learning for Underwater

Optical-Acoustic Hybrid Wireless Sensor Networks[J].

IEEE Sensors Journal, 2022,22(1): 1041-1050.

https://doi.org/10.1109/JSEN.2021.3128594

El-Banna A A A, Wu K. Machine Learning Modeling for

IoUT Networks: Internet of Underwater Things[M].

Springer Nature, 2021. https://doi.org/10.1007/978-

3-030-68567-6

Variable Rate Image Compression Based Adaptive Data Transfer Algorithm for Underwater Wireless Sensor Networks