Early Disease Detection in Apple and Grape Using Deep Learning on a Smart-Camera.

G. Polder; P.M. Blok; J.T. van Daalen; J.A. Peller; Nikos Mylonas

Early Disease Detection in Apple and Grape Using Deep Learning on a Smart-Camera.

G. Polder^*, P.M. Blok, J.T. van Daalen, J.A. Peller, Nikos Mylonas

^*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceeding › Conference paper › Academic › peer-review

Abstract

Downy mildew (Plasmopara viticola) and apple scab (Venturia inaequalis), are endemic diseases that affect crops worldwide. The diseases can cause severe losses in grapes, and apples, when it is not detected in an early stage. The EU Horizon 2020 OPTIMA project intends to address this problem by including early detection as part of an integrated pest management (IPM) system. In this research, we investigated the early detection of these two diseases using deep convolutional neural networks on RGB colour images. Detections serve as input to a Decision Support System (DSS), to precisely locate and quantify the infection, so that appropriate plant protection product, dose, timing, and location can be recommended.
A YOLOv5 object detection algorithm was developed that detected the apple scab and downy mildew spots on the apple and grape leaves. We specifically used the YOLOv5-small algorithm that worked on 224x224 pixel RGB images.
Because the DSS and the sprayer can only work with one decision per image, we combined the detections of YOLOv5 into a single decision for each image. For real-time implementation on the sprayer a smart-camera with RGB sensor and NVIDIA Jetson TX2 edge-processing unit was used. TensorRT was used to speed-up the image analysis. The average inference time for acquiring a 1.9 MP image (1600x1200 pixels) and doing the deeplearning analysis was 0.15 seconds for YOLOv5s on TensorRT. Without TensorRT this inference times was 0.20 seconds. Results showed a false detection rate of 10% for apple scab and 8.0% for downy
mildew. The accuracy of YOLOv5 was 90.0% on apple scab and 93.8% on downy mildew

Original language	English
Title of host publication	Proceedings of the European Conference on Agricultural Engineering AgEng 2021
Editors	J.C. Barbosa, L.L. Silva, P. Lourenço, A. Sousa, J.R. Silva, V.F. Cruz, F. Baptista
Publisher	EurAgEng, Cranfield
Pages	51-56
Publication status	Published - 4 Jul 2021
Event	European Conference on Agricultural Engineering AgEng 2021 - Evora, Portugal Duration: 4 Jul 2021 → 8 Jul 2021

Conference

Conference	European Conference on Agricultural Engineering AgEng 2021
Country/Territory	Portugal
City	Evora
Period	4/07/21 → 8/07/21

Access to Document

https://edepot.wur.nl/564505

Cite this

Polder, G., Blok, P. M., van Daalen, J. T., Peller, J. A., & Mylonas, N. (2021). Early Disease Detection in Apple and Grape Using Deep Learning on a Smart-Camera. In J. C. Barbosa, L. L. Silva, P. Lourenço, A. Sousa, J. R. Silva, V. F. Cruz, & F. Baptista (Eds.), Proceedings of the European Conference on Agricultural Engineering AgEng 2021 (pp. 51-56). EurAgEng, Cranfield. https://edepot.wur.nl/564505

@inproceedings{3382a1860ab44b42847428184e37b58f,

title = "Early Disease Detection in Apple and Grape Using Deep Learning on a Smart-Camera.",

abstract = "Downy mildew (Plasmopara viticola) and apple scab (Venturia inaequalis), are endemic diseases that affect crops worldwide. The diseases can cause severe losses in grapes, and apples, when it is not detected in an early stage. The EU Horizon 2020 OPTIMA project intends to address this problem by including early detection as part of an integrated pest management (IPM) system. In this research, we investigated the early detection of these two diseases using deep convolutional neural networks on RGB colour images. Detections serve as input to a Decision Support System (DSS), to precisely locate and quantify the infection, so that appropriate plant protection product, dose, timing, and location can be recommended.A YOLOv5 object detection algorithm was developed that detected the apple scab and downy mildew spots on the apple and grape leaves. We specifically used the YOLOv5-small algorithm that worked on 224x224 pixel RGB images.Because the DSS and the sprayer can only work with one decision per image, we combined the detections of YOLOv5 into a single decision for each image. For real-time implementation on the sprayer a smart-camera with RGB sensor and NVIDIA Jetson TX2 edge-processing unit was used. TensorRT was used to speed-up the image analysis. The average inference time for acquiring a 1.9 MP image (1600x1200 pixels) and doing the deeplearning analysis was 0.15 seconds for YOLOv5s on TensorRT. Without TensorRT this inference times was 0.20 seconds. Results showed a false detection rate of 10% for apple scab and 8.0% for downymildew. The accuracy of YOLOv5 was 90.0% on apple scab and 93.8% on downy mildew",

author = "G. Polder and P.M. Blok and {van Daalen}, J.T. and J.A. Peller and Nikos Mylonas",

year = "2021",

month = jul,

day = "4",

language = "English",

pages = "51--56",

editor = "J.C. Barbosa and L.L. Silva and P. Louren{\c c}o and A. Sousa and J.R. Silva and V.F. Cruz and F. Baptista",

booktitle = "Proceedings of the European Conference on Agricultural Engineering AgEng 2021",

publisher = "EurAgEng, Cranfield",

address = "United Kingdom",

note = "European Conference on Agricultural Engineering AgEng 2021 ; Conference date: 04-07-2021 Through 08-07-2021",

}

Polder, G, Blok, PM, van Daalen, JT , Peller, JA & Mylonas, N 2021, Early Disease Detection in Apple and Grape Using Deep Learning on a Smart-Camera. in JC Barbosa, LL Silva, P Lourenço, A Sousa, JR Silva, VF Cruz & F Baptista (eds), Proceedings of the European Conference on Agricultural Engineering AgEng 2021 . EurAgEng, Cranfield, pp. 51-56, European Conference on Agricultural Engineering AgEng 2021, Evora, Portugal, 4/07/21. <https://edepot.wur.nl/564505>

Early Disease Detection in Apple and Grape Using Deep Learning on a Smart-Camera. / Polder, G.; Blok, P.M.; van Daalen, J.T. et al.
Proceedings of the European Conference on Agricultural Engineering AgEng 2021 . ed. / J.C. Barbosa; L.L. Silva; P. Lourenço; A. Sousa; J.R. Silva; V.F. Cruz; F. Baptista. EurAgEng, Cranfield, 2021. p. 51-56.

Research output: Chapter in Book/Report/Conference proceeding › Conference paper › Academic › peer-review

TY - GEN

T1 - Early Disease Detection in Apple and Grape Using Deep Learning on a Smart-Camera.

AU - Polder, G.

AU - Blok, P.M.

AU - van Daalen, J.T.

AU - Peller, J.A.

AU - Mylonas, Nikos

PY - 2021/7/4

Y1 - 2021/7/4

N2 - Downy mildew (Plasmopara viticola) and apple scab (Venturia inaequalis), are endemic diseases that affect crops worldwide. The diseases can cause severe losses in grapes, and apples, when it is not detected in an early stage. The EU Horizon 2020 OPTIMA project intends to address this problem by including early detection as part of an integrated pest management (IPM) system. In this research, we investigated the early detection of these two diseases using deep convolutional neural networks on RGB colour images. Detections serve as input to a Decision Support System (DSS), to precisely locate and quantify the infection, so that appropriate plant protection product, dose, timing, and location can be recommended.A YOLOv5 object detection algorithm was developed that detected the apple scab and downy mildew spots on the apple and grape leaves. We specifically used the YOLOv5-small algorithm that worked on 224x224 pixel RGB images.Because the DSS and the sprayer can only work with one decision per image, we combined the detections of YOLOv5 into a single decision for each image. For real-time implementation on the sprayer a smart-camera with RGB sensor and NVIDIA Jetson TX2 edge-processing unit was used. TensorRT was used to speed-up the image analysis. The average inference time for acquiring a 1.9 MP image (1600x1200 pixels) and doing the deeplearning analysis was 0.15 seconds for YOLOv5s on TensorRT. Without TensorRT this inference times was 0.20 seconds. Results showed a false detection rate of 10% for apple scab and 8.0% for downymildew. The accuracy of YOLOv5 was 90.0% on apple scab and 93.8% on downy mildew

AB - Downy mildew (Plasmopara viticola) and apple scab (Venturia inaequalis), are endemic diseases that affect crops worldwide. The diseases can cause severe losses in grapes, and apples, when it is not detected in an early stage. The EU Horizon 2020 OPTIMA project intends to address this problem by including early detection as part of an integrated pest management (IPM) system. In this research, we investigated the early detection of these two diseases using deep convolutional neural networks on RGB colour images. Detections serve as input to a Decision Support System (DSS), to precisely locate and quantify the infection, so that appropriate plant protection product, dose, timing, and location can be recommended.A YOLOv5 object detection algorithm was developed that detected the apple scab and downy mildew spots on the apple and grape leaves. We specifically used the YOLOv5-small algorithm that worked on 224x224 pixel RGB images.Because the DSS and the sprayer can only work with one decision per image, we combined the detections of YOLOv5 into a single decision for each image. For real-time implementation on the sprayer a smart-camera with RGB sensor and NVIDIA Jetson TX2 edge-processing unit was used. TensorRT was used to speed-up the image analysis. The average inference time for acquiring a 1.9 MP image (1600x1200 pixels) and doing the deeplearning analysis was 0.15 seconds for YOLOv5s on TensorRT. Without TensorRT this inference times was 0.20 seconds. Results showed a false detection rate of 10% for apple scab and 8.0% for downymildew. The accuracy of YOLOv5 was 90.0% on apple scab and 93.8% on downy mildew

M3 - Conference paper

SP - 51

EP - 56

BT - Proceedings of the European Conference on Agricultural Engineering AgEng 2021

A2 - Barbosa, J.C.

A2 - Silva, L.L.

A2 - Lourenço, P.

A2 - Sousa, A.

A2 - Silva, J.R.

A2 - Cruz, V.F.

A2 - Baptista, F.

PB - EurAgEng, Cranfield

T2 - European Conference on Agricultural Engineering AgEng 2021

Y2 - 4 July 2021 through 8 July 2021

ER -

Early Disease Detection in Apple and Grape Using Deep Learning on a Smart-Camera.

Abstract

Conference

Access to Document

Fingerprint

Cite this