|
Abstract Aiming at the problems of high misidentification rate and low efficiency of manual identification in the actual environment of mushroom production and sales，21 types of common edible mushroom images of nationals using different devices and ways were collected and a dataset was established．A mushroom detection and classification model MR-YOLO was proposed on the improvement of YOLOv5s，which improves the performance of the algorithm model under the circumstance of maintaining the depth and width of the network unchanged．The network structure of the native YOLOv5s algorithm was modified，and the convolutional block attention module（CBAM） attention mechanism was introduced to ensure that the model can improve the recognition precision and accuracy while saving parameters and computation．In addition，spatial pyramid pooling cross stage paritial connection （SPPCSPC） was used in the model pooling layer to enable the model to obtain different sensory fields during the feature extraction operation，which can effectively solve the negative impacts brought by the significant differences in the size of the image targets．The C2f module was used in the model convolution layer to ensure that the model can obtain a richer information flow while being lightweight． Experimental results show that MR-YOLO improves the recognition accuracy by 4. 1% compared with the official native YOLOv5s．And the edge loss，classification loss，and confidence loss are reduced by 8.0%，16.4%，and 9.7%，respectively．It proves that the MR-yolo model has stronger performance compared with YOLOv5s.
|
|
|
|
|
|
2025, Vol. 38 
