A Lightweight YOLOv8 Tomato Detection Algorithm Combining Feature Enhancement and Attention

A tomato automatic detection method based on an improved YOLOv8s model is proposed to address the low automation level in harvesting agriculture. The provides technical support for and classification of tomatoes agricultural production activities. has three key components. Firstly, depthwise separable convolution (DSConv) technique replaces ordinary convolution, which reduces computational complexity by generating a large number feature maps with small amount calculation. Secondly, dual-path attention gate module (DPAG) designed improve model’s precision complex environments enhancing network’s ability distinguish between background. Thirdly, enhancement (FEM) added highlight target details, prevent loss effective features, precision. We built, trained, tested dataset, included 3098 images 3 classes. algorithm’s performance was evaluated comparison SSD, faster R-CNN, YOLOv4, YOLOv5, YOLOv7 algorithms. Precision, recall rate, mAP (mean average precision) were used evaluation. test results show that network lower 93.4% this dataset. This improvement 1.5% increase compared before improvement. increased 2%, rate 0.8%. Moreover, algorithm significantly reduced size from 22 M 16 M, while achieving speed 138.8 FPS, satisfies real-time requirement. strikes balance precision, enabling it meet agriculture’s requirements. research paper will provide picking robot ensure fast accurate operation robot.