Lightweight Urine Sediment Image Recognition Network Based on Deep Separable Residual Structure

The size of urine sediment image is small, different categories are easy to be confused, and feature extraction is difficult. This paper proposes an automatic recognition method of urine sediment images based on hourglass residual structure and super-resolution image reconstruction. First, annotate and preprocess the urine sediment image to generate a urine sediment data set. Then, the super-resolution reconstruction technology is used to reconstruct the small-size urine sediment image to adapt to the input of the deep learning model. Finally, an hourglass residual network is constructed to automatically extract the features of the urine sediment image to realize the classification and recognition of the urine sediment image. The experimental results show that the overall accuracy of the method for the recognition of 13 kinds of urine sediment images can reach 99.05%. This method is lightweight enough while maintaining the depth of the network. The number of parameters is 0.73M, which is conducive to porting to mobile devices. This paper proposes a new intelligent recognition method for urine sediment images, which has a good prospect for engineering applications.