Research on knee joint image classification based on improved CNN
Extracting effective information from regions is crucial for the diagnosis of knee joint magnetic resonance imaging(MRI). To extract the effective details from MRI images,a deep learning classification model that combines attention mechanisms and upsampling fusion is proposed. Firstly,useful features are enhanced and irrelevant features are suppressed through an improved channel attention mechanism. Then,the feature pyramid network(FPN)is improved through an upsampling connection mechanism,which mitigates the loss of high-level features during the upsampling process and fuses multiscale features. Finally,the detailed pooling module in MRPyrNet is employed to conduct detailed analysis on the output feature map at different scales,thereby enhancing the ability of model to capture low-level detail features. Experimental results on the MRNet dataset demonstrate that the proposed method has more advantages compared to other knee joint MRI image classification methods in terms of comprehensive performance.
万方数据
维普
