In order to explore the application of hyperspectral technology in the pathological diagnosis of gastric cancer, we combine hyperspectral imaging and microscopy to acquire hyperspectral images of gastric slices. According to the difference of spectral characteristics between gastric cancer tissue and normal gastric tissue in the wavelength of 410-910 nm, we propose a classification method based on convolutional neural network (CNN). The original spectrum is preprocessed by S-G smoothing and the first order derivative. We establish the optimal network structure and parameters by analyzing the spectral data characteristics and the classification efficiency. Experimental results show that the classification accuracy of cancerous and normal gastric tissues is 96.53%, the sensitivity and specificity of distinguishing gastric carcinoma reach 94.29% and 97.14%, respectively. Compared with shallow learning methods, the CNN model can fully extract the deep spectral characteristics of cancerous tissues and effectively prevent over-fitting. The method of deep learning combined with micro-hyperspectral imaging can also provide a new idea for the medical pathology research.