Apple Detection With FasterNet-based Lightweight YOLOv5 Algorithm

Introduction: Apple detection is an essential part of the apple-picking robot. In order to address the issues of speed and accuracy in apple detection, a lightweight YOLOv5 network with FasterNet (called FN-YOLOv5) is presented. Methods: The backbone network of the FN-YOLOv5 uses the FasterNet to reduce the large number of convolution kernels and channels, effectively balancing detection accuracy and speed. Then, the head network utilizes the ECIoU loss function. It is obtained by adding a border corner loss detection function to the CIoU for describing the deviation loss of the overlap area. Results: Therefore, the FN-YOLOv5 can obtain a balance between accuracy and speed. Discussion: Finally, the FN-YOLOv5 is compared with other classical algorithms on MS COCO2017 and self-built data-set. It is shown that the FN-YOLOv5 performs better than the other classical algorithms in terms of mean average precision, parameters, GFLOPs, FPS, Precision, Recall, and F1. Conclusion: The proposed algorithm preforms well in complex situations. The patented technology will be applied in the future.