基于改进ShuffleNet V2的板栗智能分级方法研究OA

To achieve accurate detection of appearance defects in Chinese chestnuts,the existing ShuffleNet V2 network was improved.By introducing the Convolutional Block Attention Module(CBAM),the model's ability to extract subtle defect features such as mold spots and insect holes was enhanced.Meanwhile,the Focal Loss function was adopted to alleviate the training bias caused by the imbalance between positive and negative samples.Experimental results showed that the CBAM module effectively improves the model's sensitivity to identifying defect regions,and Focal Loss strengthens the focus on hard-to-classify samples,increasing the recall rate of defect-level Chinese chestnut recognition from 85.6%to 91.8%.The proposed model maintains high classification accuracy while possessing excellent lightweight characteristics,with a parameter count of only 3.6 M and an inference time of 3.2 ms per image.