Strawberry Disease Detection in Precision Agriculture

Aguirre Santiago, Leonardo Solaque and Alexandra Velasco

Department of Engineering, Universidad Militar Nueva Granada, Bogot

a, Colombia

Keywords:

Precision Agriculture, Object Detection, Deep Learning, Crops Disease, Strawberry Crops.

Abstract:

Crop disease detection in precision agriculture has an important impact on farming, improving production, and

reducing economic losses. This is why some efforts have been done in this direction. This paper compares 4

object detection algorithms based on deep learning to detect diseases in strawberry crops. Here, we present a

step towards detecting the most common diseases to prevent economical losses. The main purpose is to detect

mainly three diseases of the strawberry crops, i.e. Botrytis cinerea, Leaf scorch, and Powdery mildew, to take

further actions if the crops are unhealthy. We have chosen these three diseases because these are frequent

and unpredictable issues, and the risk of infection is high. For this, we trained four algorithms, two based on

Single Shot MultiBox Detector and two based on EfﬁcientDet algorithm. We focus the analysis on the two

best results based on the mean average precision. We have used Google colab for training, then a Core i5 host

computer and an Nvidia Jetson nano were used for testing. We have achieved a detection network with a mean

average precision of 81% in the best case, in detecting the three proposed classes. While using an NVIDIA

Jetson nano, the accuracy increases up to 86% due to the dedicated GPU that processes Convolutional Neural

Networks(CNN).

1 INTRODUCTION

Precision agriculture has recently gained much atten-

tion due to the increasing needs of the population

around the world. There are several applications, such

as (Torky and Hassanein, 2020),(Srivastava et al.,

2019), and (Priya and Ramesh, 2020), where tech-

nologies as Blockchain or Internet of things are

brought into the agricultural ﬁeld. For example, in

((Klerkx et al., 2019) and (Lezoche et al., 2020)), cur-

rent applications that involve computer vision are ad-

dressed. Moreover, machine learning is used in ap-

plications as classiﬁcation (e.g. (

Umit Atila et al.,

2021),(Mathew et al., 2020), (Chouhan et al., 2020))

and disease detection (e.g. (Gomez Selvaraj et al.,

2020), (Mojjada et al., 2020), (Chen et al., 2020)).

Several machine learning techniques for object

detection have also been developed. For example, a

method based on deep convolution neural networks,

released in 2014 is the Region-based Convolutional

Network (R-CNN) (Girshick et al., 2013). Since then,

there have been improvements to this technique, e.g.

Fast R-CNN (Girshick, 2015), and Faster R-CNN.

(Ren et al., 2015). Other detection networks, such as

YOLO (YouOnlyLookOnce) (Redmon and Farhadi,

2017) and EfﬁcientDet (Tan et al., 2020) can also be

used for object detection. For further information on

this topic, the reader is encouraged to review (Wu

et al., 2020).

Precision agriculture is a highly growing technol-

ogy that aims to bring technology into farming pro-

cess. Crops suffer from various diseases that need to

be controlled, to prevent other problems that at their

time led to money looses. There are deﬁned ways in

agriculture to control or prevent diseases, e.g. apply-

ing different products, pruning, and so on. In ﬁelds

like fruit farming, to make decisions regarding dis-

eases prevention, e.g. fungicide application in a pre-

ventive way, information about the crops is required,

including data from the leaves and the fruit. This in-

formation can be obtained from RGB images. There-

fore, in this paper, we address the problem of object

detection using artiﬁcial intelligence, as a method to

detect diseases in strawberry crops. We compare 4

object detection algorithms, analyzing the ones that

achieve the highest mean average precision (mAP).

The analysis allows to choose the one that performs

the task more accurately, regarding the mAP criterion.

In the literature, there are different solutions to the

problem of crops disease detection in precision agri-

culture, with different targets. For example, regard-

ing machine learning techniques for classiﬁcation, in

Santiago, A., Solaque, L. and Velasco, A.

Strawberry Disease Detection in Precision Agriculture.

DOI: 10.5220/0010616405370544

In Proceedings of the 18th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2021), pages 537-544

ISBN: 978-989-758-522-7

537

(Mathew et al., 2020), authors propose a methodology

for the classiﬁcation of three important foliar diseases

in the banana crop, using texture features as ellipti-

cal local binary pattern. This method has a great ac-

curacy in classifying the diseases. However, it does

not give information about the location of the disease.

Moreover, in (Park et al., 2017), the authors propose a

deep learning mechanism to diagnose and predict dis-

eases in strawberry’s leaf and fruit. In this case, the

image has to be taken close to the fruit, and then it

has to be processed in a different engine. Regarding

leaf diseases classiﬁcation, in (Chouhan et al., 2020)

a computer vision methodology to automate the dis-

ease diagnosis of Jatropha Curcas (Huang et al., 2020)

is proposed with high classiﬁcation accuracy. In this

case, the authors have identiﬁed that increasing the

segmentation performance and using a Deep neural

network for the classiﬁcation task would yield to bet-

ter results, due to the difference between the use of

artiﬁcial intelligence algorithms there is no room for

comparison, because the authors use AI to classify

whereas in this article, it is to detect.

There are other solutions to the scope problem that

do not use artiﬁcial intelligence, but they use digi-

tal image processing. For example, in (Sheikh et al.,

2019) the authors implemented an image processing

algorithm and deep learning methods on images of

crops with diseases, to help the farmers to cultivate

and reducing the diseases. The algorithm has a great

accuracy, but it does not provide the location of the

disease in the plant. Moreover, the work is only fo-

cused in detecting problems in the leaves. Regard-

ing only image processing, for instance in (Prakash

et al., 2017), the authors propose a framework that

includes image preprocessing, Segmentation using

clustering feature extraction by statistical Gray-Level

Co-Occurrence Matrix (GLCM) & Classiﬁcation of

diseases using Suport Vector Machines(SVM). This

algorithm could be implemented in different plant

species with few changes.

Object detection can be applied to solve other

tasks such as position tracking, which is useful in

many ﬁelds such as autonomous driving. For ex-

ample, this problem can be solved with deep learn-

ing algorithms, training a detection Network such as

YOLO and analyzing the changes in the generated

detection. In (Ciaparrone et al., 2020), the authors

provide a survey on Deep Learning Models that solve

the task of Multiple Object Tracking on single-camera

videos, comparing several models and demonstrating

that Deep Learning algorithms are as effective solving

this task as solutions such as LiDar and depth images.

In this paper, we show a comparison of 4 algo-

rithms to detect strawberry diseases, i.e. powdery

mildew, botrytis cinerea, leaf scorch, immature straw-

berry, and healthy strawberries. At the end, we deter-

mine the best detection algorithm based on the best

mAP. For this, we trained two deep learning based

object detection algorithms in a custom dataset. Fig.1

illustrates the implemented strategy. First, we cre-

ate a custom dataset for the required classes using

some data sources. Then, we have an image pro-

cessing stage where we normalize the images, for this

we use the OpenCV library. We generate an infer-

ence graph that can be used in other computers with

lower speciﬁcations than the host computer. This was

done employing the Tensorﬂow 2.0, an object detec-

tion API, as deep learning framework. The model was

trained in Google colab with Intel(R) Xeon(R) CPU

@ 2.20GHz and 16GB NVIDIA Tesla T4 GPU. How-

ever, the trained models were tested in an Intel Core

i5-7200 with 2 GB NVIDIA Geforce MX940 GPU.

We aim to detect 3 common diseases and two

growing states in the strawberry crops, with the idea

of taking corrective actions using the CERES agricul-

tural robot. We address the strategy to detect the dis-

eases and the growing states, while the decision mak-

ing algorithm is not part of this paper.

In section 2 we present the complete strategy for

object detection, showing how we trained the ob-

ject detection algorithms and the image processing as

well. Section 2.1 presents how and why we selected

the proposed classes. We analyze the results in sec-

tion 3, and we give some conclusions and recommen-

dations in section 4.

2 METHODS

In this section, we describe the strategy used to de-

tect strawberry diseases. To do this, we trained the

EfﬁcientDet detection network model with efﬁcient-

Net, (Tan et al., 2020), Single Shot MultiBox Detec-

tor(SSD) with Resnet 50 (Liu et al., 2015), SSD with

mobilenet V2. Every detection network uses differ-

ent methods to solve the detection process, Efﬁcient-

Det, works with EfﬁcientNet (Tan and Le, 2019) as

the backbone network, BiFPN as the feature network,

and shared class/box prediction network. SSD works

with Resnet 50 (He et al., 2015) as backbone. It is

a feed-forward convolutional network that produces a

ﬁxed-size collectionThe more accurate the model is

the of bounding boxes and scores for the presence of

object class instances in those boxes, followed by a

non-maximum suppression step to produce the ﬁnal

detections.

For the purpose of our work, we trained every de-

tection network with a custom dataset created with

ICINCO 2021 - 18th International Conference on Informatics in Control, Automation and Robotics

538

PlantDoc dataset (Singh et al., 2019), Strawberry

dataset (P

erez-Borrero et al., 2020) and a custom dat-

set created from Google images. The former dataset

contains 450 images. The hole dataset has a complex

background with a remarkable light change.

Figure 1: Implemented strategy and project pipeline.

2.1 Data Acquisition

In this paper we aim to detect some of the most com-

mon strawberry diseases, as well as immature and

healthy strawberries. The selected diseases in this

project were:

• Powdery Mildew: This is a fungal disease which

affects a wide range of ﬁeld crops, trees, shrubs,

vines, ﬂowers, vegetables, fruits, grasses, and

weeds. The most common symptom in the in-

fected plants is the presence of white powdery

spots on leaves and stems. The lower leaves are

the most affected part, but the disease can appear

in any part of the plant. Additionally, powdery

mildew, causes powdery growth on the surface of

young shoots, leaves, ﬂowers, and fruits. Pow-

dery mildew is caused by many variants of fun-

gal species in the genera Erysiphe, Microsphaera,

Phyllactinia, Podosphaera, Sphaerotheca, and

Uncinula(Carisse and Fall, 2021).

• Botrytis Cinerea: It is a fungal pathogen that

causes grey mould mainly in the fruits. This

pathogen affects the fruits in the ﬁeld, storage,

transport and market. The presence of grey mould

is the most common reason for fruit rejection by

growers, shippers and consumers, leading to sig-

niﬁcant economic losses (Petrasch et al., 2019).

• Leaf Scorch: It is caused by a fungal infection

which affects the foliage of strawberry plantings.

The fungal species responsible are called Diplo-

carpon earliana. Strawberries with leaf scorch

may ﬁrst show signs of issue with the develop-

ment of small purplish blemishes that occur on the

topside of leaves. If the disease is allowed to ad-

vance, the spots will grow larger and darker. In

the worst cases, those spots may even cover entire

portions of the crop, including plant leaves and

cause them to completely dry and fall from the

plant.

• Inmature Strawberry: this includes fruit light in

mass, stunted or distinctly rubbery in texture. In

the case of the strawberry, it is also caracterized

by a green color.

• Healthy Strawberry: Fruit in great conditions.

For further information of the diseases and the straw-

berry’s growing, the reader is referred to (Hancock

et al., 2008) and (Vanti et al., 2021).

The proposed dataset consists of 450 images split-

ted in a training set with 300 positive images. The

test set has 100 positive images and the validation test

set consists of 50 images. For each class, there is an

amount of 90 images distributed in 60 images for the

training set, 10 images for the validation set and 20

images for the test set. However, the number of anno-

tations per class is:

• Powdery Mildew: 188 annotations.

• Botrytis Cinerea: 179 annotations.

• Leaf Scorch: 174 annotations.

• Inmature Strawberry: 193 annotations.

• Strawberry: 186 annotations.

2.2 Object Detection

Object detection algorithms have many applications

such as autonomous driving. For example, companies

like Tesla, Apple, Toyota, Nissan, etc., use them to

avoid collisions during a course (Wang et al., 2020).

In this paper, we compare 4 object detection algo-

rithms and we deﬁne which one ﬁts better with our re-

quirements to be implemented in the CERES agricul-

tural robot (Santiago. et al., 2020), based on the mAP.

For this, we detect the 5 classes mentioned before.

The proposed architecture is shown in Fig. 2. The

Strawberry Disease Detection in Precision Agriculture

539

Figure 2: Proposed architecture for the research.

process consists of an image acquisition step; then,

the detection algorithm returns the bounding boxes

with the detection, and the detected class. In this sec-

tion, we show the training process for each detection

network and the image pre-processing stage, Fig. 1 il-

lustrates the ﬂowchart of the strategy proposed in this

paper.

2.3 Models Training

The ﬁrst stage of the training process includes im-

age prepossessing and normalization, as shown in Fig.

1. This is done in OpenCV, where we apply trans-

formations such as random horizontal ﬂip, crop and

re-scale, to get more images for the process. Then,

the re-sizing process for our images varies depend-

ing of the detection algorithm, e.g for EfﬁcientDet,

the image input shape is 512x512 pixels. After that,

we normalize the images between 0-1, and ﬁnally, we

change the image format to RGB, because OpenCV

works with BGR image format. In this way, we have

data ready to train the object detection models.

The training process is shown in Fig. 3. We tested

two object detection deep learning based algorithms

with different backbone CNN. The training process

consists of ﬁve steps. First, the generation of the

record ﬁles, which is done by running some python

scripts after labeling every single image in the dataset.

This is done because the data are in ’xml’ format, but

it is needed in ’recod’ data format. The second step

is to select the backbone Convolutional Neural Net-

work (CNN) for the detection algorithm. In table 1

we show the CNN for each iteration. The fourth step

includes the setting of the training options, which are

also shown in Table 1. Those hyperparameters were

chosen to avoiding overﬁtting of the obtained results.

Finally, a label map for the classes to start training the

algorithms is required.

Additionally, the batch size of each training pro-

cess is limited by the hardware capacity. Our host

computer reduced the batch size of 1, while in Google

colab we were able to use a batch between 8 to 16, due

the Testa T4 GPUs provided.

3 RESULTS AND ANALYSIS

In this section we present and explain the results of

training the two algorithms, i.e. EfﬁcientDet-D0 and

SSD-Resnet50. These two algorithms give us the best

practical results based on the mean average precision

(mAP).

Testing was carried out with the test set, which

contains 100 images. However, the hyperparameters

change in each training iteration. In table 1 we show

the number of training iterations done for each algo-

rithm. Every model iteration was tested in a 50 im-

ages validation set. During the experimentation pro-

cess, we used a 60% threshold during the detection

task.

The tests consisted on the detection process of an

image test set. The results of the detector, the scores,

Table 1: Training parameters for each detection algorithm.

Train iterations Algorithm Backbone CNN Iterations Input shape Batch size

5 EfﬁcientDet-d0 EfﬁcientNet 20000 512x512 16

4 SSD Resnet 50 15000 640x640 8

2 EfﬁcientDet-d3 EfﬁcientNet 10000 768x768 8

2 SSD Mobilenet 10000 320x320 16

ICINCO 2021 - 18th International Conference on Informatics in Control, Automation and Robotics

540

Figure 3: Training process for the detection algorithms.

and the bounding boxes per image are compared with

the ground truth proposed for the test set. The ground

truth is a table that contains the information about the

location of the image on the computer where the test

is carried out, and the bboxes of each image. The

trained models were tested in a NVIDIA JETSON

Nano ARM

 Cortex

 -A57 MPCore with Maxwell

128 GPU Uand in an Intel Core i5-7200 with 2 GB

NVIDIA Geforce MX940 GPU.

The average precision is deﬁned as the average of

the precision scores after each true positive. The mAP

compares the ground-truth bounding box with the de-

tected box. The more accurate the model, the higher

the mAP value. In Table 2, the Average precision re-

sults per class are shown as well as the mAP of each

detection network.

3.1 EfﬁcientDet-D0

The training process with the dataset proposed in sec-

tion 2.1 lasted 175 minutes, using Intel(R) Xeon(R)

CPU @ 2.20GHz and 16GB NVIDIA Tesla T4 GPU

provided by Google Colab. In Fig. 4 the training loss

by the epochs is shown. The ﬁnal training iteration

process consisted of 15000 iterations with 16 Batch

size. According to the amount of data used to train

Figure 4: Training loss graph for EfﬁcientDet - EfﬁcientNet

D0.

the detector, we achieved a mAP of 81% in the Core

i5 Computer, and 86% with a minimum error of 0.034

in the NVIDIA Jetson Nano which reﬂects the action

of the cuBlas and cuDNN libraries which increases

the detection accuracy in 5% on the Jetson with re-

spect to the Core i5 computer. The inference time is

about 300 ms. This was the best result obtained during

the trials. The training iterations are shown in table 1.

Table 2: Detection rates.

Strawberry Inmature Straw Powdery Mildew Botrytis Leaft Scoarch MAP

EfﬁcientDet-D0 92 88 76 89 86 86,2

EfﬁcientDet-D3 95 92 64 72 80 80,6

SSD-Resnet50 90 85 78 81 80 82,8

SSD-Mobilenet 76 78 68 80 74 75,2

Strawberry Disease Detection in Precision Agriculture

541

3.2 SSD-Resnet50

The training process with the dataset proposed in sec-

tion 2.1 lasted 200 minutes, using Intel(R) Xeon(R)

CPU @ 2.20GHz and 16GB NVIDIA Tesla T4 GPU

provided by Google Colab. In Fig. 4 the training loss

by the epochs is shown. The ﬁnal training iteration

process consisted on 20000 iterations with 8 Batch

size.

Figure 5: Training loss graph for SSD - resnet 50.

According to the amount of data used to train the

detector, we achieved a mAP of 77% in the Core i5

Computer, and 83% with a minimum error of 0.034

in the NVIDIA Jetson Nano , which evidences that

the cuBlas and cuDNN libraries increase the detection

accuracy in 6% on the Jetson with respect to the Core

i5 computer. The inference time is about 215 ms.

In Fig. 6 we show a set of output images of the

best model, that in our case, for the proposed task was

EfﬁcientDet-D0.

4 CONCLUDING REMARKS

In this paper, we chose a strategy to detect diseases

in strawberry crops. To do this, we compared 4 ob-

ject detection algorithms based on deep learning, and

we presented the best two algorithms regarding their

mean average precision (mAP). With the chosen strat-

egy we detect three diseases of the strawberry crops,

and two growing states, i.e. immature strawberry and

healthy strawberry.

For this work we used Google colab. Then a

Core i5 host computer and a Nvidia Jetson nano were

used for testing. We have achieved a detection net-

work with a mAP of 81% in detecting the three pro-

posed diseases (classes). While using a NVIDIA Jet-

Figure 6: Results of detecting the proposed classes.

ICINCO 2021 - 18th International Conference on Informatics in Control, Automation and Robotics

542

son nano, the accuracy increases up to 86% due to

the dedicated GPU that process Convolutional Neural

Networks(CNN).

The use of better detection datasets would in-

crease the obtained mAP, mainly in the detection of

powdery mildew disease, where the average precision

was of 75% in the EfﬁcientDet-D0 best train iteration.

Moreover, an optimization step regarding the Nvidia

Jetson nano with TensorRT would decrease the infer-

ence time. We will address this in our future work for

the implementation of the algorithm in the CERES

agro-robot (Santiago. et al., 2020).

ACKNOWLEDGEMENTS

This work is funded by Universidad Militar Nueva

Granada- Vicerrectoria de Investigaciones, under re-

search grant for project INV ING 3185 “Sistema de

toma de decisiones para la aplicaci

on de medidas cor-

rectivas que ayuden a mantener la salud de un cultivo

de hortalizas utilizando un robot (CERES) dedicado a

labores de agricultura”.

REFERENCES

Carisse, O. and Fall, M. (2021). Decision trees to

forecast risks of strawberry powdery mildew caused

by podosphaera aphanis. Agriculture (Switzerland),

11(1):1–16. cited By 0.

Chen, J., Yin, H., and Zhang, D. (2020). A self-adaptive

classiﬁcation method for plant disease detection using

gmdh-logistic model. Sustainable Computing: Infor-

matics and Systems, 28:100415.

Chouhan, S., Singh, D. U., Sharma, U., and Jain, S. (2020).

Leaf disease segmentation and classiﬁcation of jat-

ropha curcas l. and pongamia pinnata l. biofuel plants

using computer vision based approaches. Measure-

ment, 171.

Ciaparrone, G., S

anchez, F. L., Tabik, S., Troiano, L., Tagli-

aferri, R., and Herrera, F. (2020). Deep learning in

video multi-object tracking: A survey. Neurocomput-

ing, 381:61 – 88.

Girshick, R. (2015). Fast r-cnn. In 2015 IEEE International

Conference on Computer Vision (ICCV), pages 1440–

1448.

Girshick, R. B., Donahue, J., Darrell, T., and Malik, J.

(2013). Rich feature hierarchies for accurate ob-

ject detection and semantic segmentation. CoRR,

abs/1311.2524.

Gomez Selvaraj, M., Vergara, A., Montenegro, F., Alonso

Ruiz, H., Safari, N., Raymaekers, D., Ocimati, W.,

Ntamwira, J., Tits, L., Omondi, A. B., and Blomme,

G. (2020). Detection of banana plants and their major

diseases through aerial images and machine learning

methods: A case study in dr congo and republic of

benin. ISPRS Journal of Photogrammetry and Remote

Sensing, 169:110 – 124.

Hancock, J., Sjulin, T., and Lobos, G. (2008). Strawberries,

volume 9781402069079. cited By 36.

He, K., Zhang, X., Ren, S., and Sun, J. (2015). Deep

residual learning for image recognition. CoRR,

abs/1512.03385.

Huang, J.-D., Wang, C.-F., Lian, C.-L., Huang, M.-Y.,

Zhang, C., and Liu, J.-Q. (2020). Isolation and identi-

ﬁcation of ﬁve new diterpenoids from jatropha curcas.

Phytochemistry Letters, 40:37–41.

Klerkx, L., Jakku, E., and Labarthe, P. (2019). A review

of social science on digital agriculture, smart farming

and agriculture 4.0: New contributions and a future

research agenda. NJAS - Wageningen Journal of Life

Sciences, 90-91:100315.

Lezoche, M., Hernandez, J. E., del Mar Eva Alemany D

ıaz,

M., Panetto, H., and Kacprzyk, J. (2020). Agri-food

4.0: A survey of the supply chains and technologies

for the future agriculture. Computers in Industry,

117:103187.

Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S. E.,

Fu, C., and Berg, A. C. (2015). SSD: single shot multi-

box detector. CoRR, abs/1512.02325.

Mathew, D., Sathish Kumar, C., and Anita Cherian, K.

(2020). Foliar fungal disease classiﬁcation in banana

plants using elliptical local binary pattern on multires-

olution dual tree complex wavelet transform domain.

Information Processing in Agriculture.

Umit Atila, Uc¸ar, M., Akyol, K., and Uc¸ar, E. (2021).

Plant leaf disease classiﬁcation using efﬁcientnet deep

learning model. Ecological Informatics, 61:101182.

Mojjada, R. K., Kiran Kumar, K., Yadav, A., and Satya Vara

Prasad, B. (2020). Detection of plant leaf disease us-

ing digital image processing. Materials Today: Pro-

ceedings.

Park, H., Eun, J., and Kim, S. (2017). Image-based dis-

ease diagnosing and predicting of the crops through

the deep learning mechanism. In 2017 International

Conference on Information and Communication Tech-

nology Convergence (ICTC), pages 129–131.

Petrasch, S., Knapp, S. J., van Kan, J. A. L., and Blanco-

Ulate, B. (2019). Grey mould of strawberry, a devas-

tating disease caused by the ubiquitous necrotrophic

fungal pathogenBotrytis cinerea. Molecular Plant

Pathology, 20(6):877–892.

Prakash, R. M., Saraswathy, G. P., Ramalakshmi, G., Man-

galeswari, K. H., and Kaviya, T. (2017). Detection

of leaf diseases and classiﬁcation using digital image

processing. In 2017 International Conference on In-

novations in Information, Embedded and Communi-

cation Systems (ICIIECS), pages 1–4.

erez-Borrero, I., Mar

ın-Santos, D., Geg

undez-Arias,

M. E., and Cort

es-Ancos, E. (2020). A fast and ac-

curate deep learning method for strawberry instance

segmentation. Computers and Electronics in Agricul-

ture, 178:105736.

Priya, R. and Ramesh, D. (2020). Ml based sustainable

precision agriculture: A future generation perspec-

Strawberry Disease Detection in Precision Agriculture

543

tive. Sustainable Computing: Informatics and Sys-

tems, 28:100439.

Redmon, J. and Farhadi, A. (2017). Yolo9000: Better,

faster, stronger. In 2017 IEEE Conference on Com-

puter Vision and Pattern Recognition (CVPR), pages

6517–6525.

Ren, S., He, K., Girshick, R., and Sun, J. (2015). Faster

r-cnn: Towards real-time object detection with region

proposal networks. In Cortes, C., Lawrence, N. D.,

Lee, D. D., Sugiyama, M., and Garnett, R., editors,

Advances in Neural Information Processing Systems

28, pages 91–99. Curran Associates, Inc.

Santiago., A., Solaque., L., and Velasco., A. (2020). Deep

learning algorithm for object detection with depth

measurement in precision agriculture. In Proceed-

ings of the 17th International Conference on Informat-

ics in Control, Automation and Robotics - Volume 1:

ICINCO,, pages 490–497. INSTICC, SciTePress.

Sheikh, M. H., Mim, T. T., Reza, M. S., and Hena, M. H.

(2019). Leaf diseases detection for commercial cul-

tivation of obsolete fruit in bangladesh using image

processing system. In 2019 8th International Confer-

ence System Modeling and Advancement in Research

Trends (SMART), pages 271–275.

Singh, D., Jain, N., Jain, P., Kayal, P., Kumawat, S., and

Batra, N. (2019). Plantdoc: A dataset for visual plant

disease detection. CoRR, abs/1911.10317.

Srivastava, K., Bhutoria, A. J., Sharma, J. K., Sinha, A.,

and Pandey, P. C. (2019). Uavs technology for the de-

velopment of gui based application for precision agri-

culture and environmental research. Remote Sensing

Applications: Society and Environment, 16:100258.

Tan, M. and Le, Q. V. (2019). Efﬁcientnet: Rethink-

ing model scaling for convolutional neural networks.

CoRR, abs/1905.11946.

Tan, M., Pang, R., and Le, Q. V. (2020). Efﬁcientdet: Scal-

able and efﬁcient object detection.

Torky, M. and Hassanein, A. E. (2020). Integrat-

ing blockchain and the internet of things in preci-

sion agriculture: Analysis, opportunities, and chal-

lenges. Computers and Electronics in Agriculture,

178:105476.

Vanti, G., Leshem, Y., and Masaphy, S. (2021). Resis-

tance response enhancement and reduction of botrytis

cinerea infection in strawberry fruit by morchella con-

ica mycelial extract. Postharvest Biology and Technol-

ogy, 175. cited By 0.

Wang, L., Fan, X., Chen, J., Cheng, J., Tan, J., and Ma, X.

(2020). 3d object detection based on sparse convolu-

tion neural network and feature fusion for autonomous

driving in smart cities. Sustainable Cities and Society,

54:102002.

Wu, X., Sahoo, D., and Hoi, S. C. (2020). Recent advances

in deep learning for object detection. Neurocomput-

ing.

ICINCO 2021 - 18th International Conference on Informatics in Control, Automation and Robotics

544