关键词: 绿橙, 目标检测, 模型轻量化, 深度学习

Abstract: The color of the green orange fruits is similar to that of the branches and leaves, making identification challenging. To solve the problems of high-precision detection and model lightweight improvement of green orange, this study selects the “green orange” from the orange base in Meishan City, Sichuan Province as the research object and proposes an improved target detection algorithm named ESN—YOLO based on YOLOv8n. First, GSConv and VoV—GSCSP modules are introduced into the Neck section of the model to significantly reduce its parameters. Then, the SimAM attention mechanism is integrated into the Neck network layer to enhance the detection performance of the model. Finally, the loss function is modified from CIoU to EIoU to improve the models precision and accuracy in target detection. The results show that the modified ESN—YOLO model achieves an accuracy of 96.9%, mAP@0.5 of 99.7%, and mAP@0.5∶0.95 of 83.6%. Compared with the original YOLOv8n model, these values represent improvements of 3.2%, 0.5%, and 3.6%, respectively. The model size of the improved algorithm is 5730 KB, and the model parameters are 2.6 MB, which are reduced by 11.4% and 13.9%, respectively. ESN—YOLO achieved an effective balance between model lightweight and target detection accuracy. The improved model is deployed on the embedded device Jetson Nano, where the detection speed for a single orange photo is 732ms, with an accuracy rate exceeding 90%. 

Key words: green orange, target detection, model lightweight, deep learning