Rice-YOLO: An Automated Insect Monitoring in Rice Storage Warehouses with the Deep Learning Model

Abstract

Grain storage is an essential component of grain supply chain management that guarantees food security within the nation. Inaccurate diagnosis of insect infestation during grain storage might lead to misinterpretation of fumigation, resulting in substantial qualitative and quantitative losses of grains. This work introduces a new deep learning model called “Rice-YOLO” (You Only Look Once) that addresses the shortcomings of existing insect detection methods. The model offers a high level of accuracy and real-time performance. This model has been optimized to accurately identify Tribolium castaneum and Rhyzopertha dominica in stored rice grains, under different background and lighting circumstances. YOLOv7 (YOLOv7 and x) and YOLOv8 (l/m/x/s/n) were the models used to train, test, and validate the insect data sets. The performance of these state-of-the-art deep learning models was assessed. YOLOv8 obtained remarkable outcomes on the Rice data set. It achieved 97.7% mean average precision (mAP) and 97.5% recall for T. castaneum, as well as a precision of 95.5%. R. dominica scored a mAP of 96.2% and a recall of 93%. The model took around 7.68 min to process and detect T. castaneum and R. dominica. The top-performing YOLOv8n model was then deployed on a laptop achieving a detection speed of 22 fps and an inference time of 6.4 ms. The findings indicated that the algorithm was rapid and effective in detecting, identifying, and quantifying insect pests in stored grains. This could facilitate the automatic identification of insects in warehouses and grain storage facilities involved in effective postharvest management.