Search

In complex scenes, factors such as tree branches and leaves occlusion, dense distribution of tomato fruits, and similarity of fruit color to the background color make it difficult to correctly identify the ripeness of the tomato fruits when harvesting them. Therefore, in this study, an improved YOLOv8 algorithm is proposed to address the problem of tomato fruit ripeness detection in complex scenarios, which is difficult to carry out accurately. The algorithm employs several technical means to improve detection accuracy and efficiency. First, Swin Transformer is used to replace the third C2f in the backbone part. The modeling of global and local information is realized through the self-attention mechanism, which improves the generalization ability and feature extraction ability of the model, thereby bringing higher detection accuracy. Secondly, the C2f convolution in the neck section is replaced with Distribution Shifting Convolution, so that the model can better process spatial information and further improve the object detection accuracy. In addition, by replacing the original CIOU loss function with the Focal–EIOU loss function, the problem of sample imbalance is solved and the detection performance of the model in complex scenarios is improved. After improvement, the mAP of the model increased by 2.3%, and the Recall increased by 6.8% on the basis of YOLOv8s, and the final mAP and Recall reached 86.9% and 82.0%, respectively. The detection speed of the improved model reaches 190.34 FPS, which meets the demand of real-time detection. The results show that the improved YOLOv8 algorithm proposed in this study exhibits excellent performance in the task of tomato ripeness detection in complex scenarios, providing important experience and guidance for tomato ripeness detection. [ABSTRACT FROM AUTHOR]