The invention relates to the technical field of image processing, and concretely relates to a gastric cancer focus detection method and device based on a convolutional neural network. The gastric cancer lesion detection method based on the convolutional neural network comprises the following steps: S1, preprocessing a general image of a gastric cancer sample to be detected; S2, performing focus target extraction and confidence analysis based on a target detection algorithm model, and outputting a focus detection result; or S3, finely segmenting and outlining the focus target based on the semantic segmentation algorithm model, and outputting a focus detection result. According to the method, the general image of the gastric cancer sample is utilized for the first time, the cancer lesion and intragastric or perigastric metastatic cancer lesion in the gastric resection specimen can be automatically positioned, meanwhile, the confidence coefficient of an analysis result is given, an examination doctor is assisted in accurately cutting a lesion part of the specimen, the cancer lesion detection efficiency is improved, and the missed diagnosis rate is reduced.