Lateral bending detection method and model based on multi-semantic segmentation network

The invention relates to a lateral bending detection method and model based on a multi-semantic segmentation network. The lateral bending detection method comprises the following steps: respectively inputting an image to be identified into a first segmentation subnet and a second segmentation subnet in a trained multi-view semantic segmentation network for semantic segmentation to obtain a first subnet segmentation graph and a second subnet segmentation graph, performing uncertainty minimization competition on the two subnet segmentation images to synthesize a segmentation prediction image; carrying out human body key part identification and key point matching on the segmented prediction image to obtain a plurality of matched key points; calculating index parameters according to a plurality of matched key points, outputting a prediction result according to the index parameters, emphasizing learning of image features of the segmented image through a plurality of subnets, extracting important features of two im