Human Stools Classification for Gastrointestinal Health based on an Improved ResNet18 Model with Dual Attention Mechanism

The human stools are directly related to the health of human gastrointestinal function. Preliminary classification of the shape and colour of stools can diagnose the health status of peoples, therefore automatic recognition of stools is the current development direction of smart toilets. Due to the difficulty in identification with complex image content, this paper proposed a convolutional neural network called StoolNet to solve the current challenges. The architecture of StoolNet is based on ResNet and contains two output branches which perform colour and shape recognition, respectively. To improve the recognition performance, the dual attention mechanism was introduced into feature extraction stage. The accuracy value of our proposed model could achieve 99.7% and 94.4% for color and shape recognition on our test set, respectively. Experimental results show that, compared with other stool classification algorithms, our method possesses better capability of category discrimination on real dataset.