Metabolic syndrome (MetS) is an increasingly common group of metabolic abnormalities, with up to one-third of American adults living with the disorder. Metabolic syndrome is closely related to diabetes and cardiovascular disease, and if not detected early, it will seriously harm human health. The morphological characteristics of human feces are the obvious manifestation of metabolic status. By identifying the morphological characteristics of feces, metabolic abnormalities can be identified and early treatment and prevention of metabolic syndrome can be realized. In the past, the examination of fecal traits was usually done by professionals with visual inspection, which was subjective and time-consuming. There were few researches on automatic recognition and classification of metabolic products by computer vision. In this paper, a deep convolutional neural network (CNN) model is proposed to monitor metabolic conditions by classifying stool images of normal and abnormal metabolism. In this paper, 210 stool samples were collected, 1890 stool photos were taken, and three deep learning models were used to achieve image block-level classification. After training and testing the dataset, ResNet-50 achieved 99.86% accuracy, and all classifiers scored above 84% F1 in all categories. In view of the high accuracy of the trained ResNet-50 and its better generalization ability, we recommend that this model be used in the classification of fecal traits.