Identifying common stored product insects using automated deep learning methods

Authors: BADGUJAR, CHETAN - Oak Ridge Institute For Science And Education (ORISE); Armstrong, Paul; Gerken, Alison; Pordesimo, Lester; Campbell, James - Jim

Submitted to: Journal of Stored Products Research
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 7/23/2023
Publication Date: 8/3/2023
Citation: Badgujar, C., Armstrong, P.R., Gerken, A.R., Pordesimo, L.O., Campbell, J.F. 2023. Identifying common stored product insects using automated deep learning methods. Journal of Stored Products Research. 103. Article 102166. https://doi.org/10.1016/j.jspr.2023.102166.
DOI: https://doi.org/10.1016/j.jspr.2023.102166

Interpretive Summary: Monitoring of stored product insect pests is a common practice for management of stored grain and helps to ensure grain quality from storage to sale. Current manual sampling and monitoring methods for insect activity in large grain storage facilities and food production facilities are time-consuming, labor-intensive, require expertise for accurate species identification, and overall, are expensive. Using artificial intelligence (AI) models to provide accurate species identification may help alleviate this bottleneck and allow more rapid interventions when pests populations are detected, resulting in reduced damage and economic losses. This study developed automated, image-based, identification of five common stored grain insect species (lesser grain borer, rusty grain beetle, red flour beetle, rice weevil, saw-toothed grain beetle) using deep learning, which is an emerging AI algorithm that is being increasingly used in agriculture. Identification was very reliable with at least 96% accuracy for all species with few misclassifications. Gradient-weighted Class Activation Mapping (Grad-CAM) was used to identify the specific insect features that were most useful for reliable species identification. Generally, differences in the shapes of the insect body and head were most significant to discern species, with differences in antennae, legs and snouts being less critical for identification. This study contributes to the overall goal of developing a camera-based system for automated monitoring of stored grain insects. This system would empower warehouse, flour mills, and other storage facilities with an inexpensive tool to autonomously identify insect species in stored product environments and could be implemented as part of a real-time monitoring system for more effective and rapid control and understanding insect behavior.

Technical Abstract: Monitoring of stored product insect pests is a common practice for post-harvest management of stored grain and helps ensure quality grain from storage to sale. Current methods of sampling and monitoring are time-consuming, labor-intensive, and expensive. Electronic monitors exist but currently do not provide accurate species identification. The objective of this study is to develop automated, image-based, identification of common stored grain insect species using deep-learning methods. Top-down images of common stored product insects were captured, and comprised the species of adult Rhyzopertha dominica, Cryptolestes ferrugineus, Tribolium castaneum, Sitophilus oryzae, and Oryzaephilus surinamensis. The deep learning-based, state-of-the-art Convolutional Neural Networks (CNN) models (ResNet-50, MobileNet-v2, DarkNet-53, and EfficientNet-b0) were fine-tuned with a transfer learning approach to classify the insect species. The model performed very well with few misclassifications, and all species were identified correctly with at least 96% accuracy. Trained networks are generally unable to explain the reasoning behind predicted identification properties and act, and are often called a “black box”. Therefore, visualization methods called Gradient-weighted Class Activation Mapping (Grad-CAM) were implemented to explore the significant features used by the network. Grad-CAMs uses heat maps to highlight the major image features that the network focused on to make insect species predictions, verifies the network’s prediction, and where the performance can be improved. This study contributes to the overall goal of developing a camera-based system for monitoring of stored grain insects. The developed system would empower warehouse, flour mills, and other storage facilities with an inexpensive tool to accurately identify insect species in stored product environments and could be implemented as part of a close to real-time monitoring system.