ECLPOD: An Extremely Compressed Lightweight Model for Pear Object Detection in Smart Agriculture

Accurate pear sorting plays a crucial role in ensuring the quality of pears and increasing sales them. In domain intelligent sorting, precise target detection is imperative. However, practical implementation faces challenges achieving adequate accuracy due to limitations computational resources embedded devices occurrence occlusion among pears. To solve this problem, we built an image acquisition system based on equipment created dataset containing 34,598 images under laboratory conditions. The was meticulously annotated using LabelImg software, resulting total 154,688 annotations for pears, stems, calyxes, defects. Furthermore, propose Extremely Compressed Lightweight Model Pear Object Detection (ECLPOD) YOLOv7’s pipeline assist task. Firstly, Hierarchical Interactive Shrinking Network (HISNet) proposed, which contributed efficient feature extraction with limited amount computation parameters. Bulk Feature Pyramid (BFP) module then proposed enhance contour information during fusion. Finally, Accuracy Compensation Strategy (ACS) improve capability model, especially identification calyces stalks experimental results indicate that ECLPOD achieves 90.1% precision (P) 85.52% mAP50 only 0.58 million parameters 1.3 GFLOPs homemade paper. Compared YOLOv7, number are compressed 1.5% 1.3%, respectively. other mainstream methods, optimal trade-off between complexity. This suggests superior these existing approaches field object assisting tasks good potential device deployment.