An Improved YOLOv5s-Based Agaricus bisporus Detection Algorithm

This study aims to improve the Agaricus bisporus detection efficiency and performance of harvesting robots in the complex environment of the mushroom growing house. Based on deep learning networks, an improved YOLOv5s algorithm was proposed for accurate A. bisporus detection. First, A. bisporus images collected in situ from the mushroom growing house were preprocessed and augmented to construct a dataset containing 810 images, which were divided into the training and test sets in the ratio of 8:2. Then, by introducing the Convolutional Block Attention Module (CBAM) into the backbone network of YOLOv5s and adopting the Mosaic image augmentation technique in training, the detection accuracy and robustness of the algorithm were improved. The experimental results showed that the improved algorithm had a recognition accuracy of 98%, a single-image processing time of 18 ms, an A. bisporus center point locating error of 0.40%, and a diameter measuring error of 1.08%. Compared with YOLOv5s and YOLOv7, the YOLOv5s-CBAM has better performance in recognition accuracy, center positioning, and diameter measurement. Therefore, the proposed algorithm is capable of accurate A. bisporus detection in the complex environment of the mushroom growing house.