Deep convolution neural network is a very effective method in the field of computer vision. With the rapid growth of image data and the increasing popularity of intelligent devices, it is required to understand the content of the image quickly and accurately, and automatically segment and recognize the object in the image. The task of image segmentation is to detect whether a certain kind of object is included in a given image, mark the object category of each pixel in the image, describe the boundary of each object, and finally obtain a segmentation image with pixel semantic annotation. The detection and segmentation of the object in the image is very important for the development of computer vision, and also has high practical value in the actual engineering application. At present, many excellent research results about the image segmentation algorithm have been published at home and abroad, but when it is applied to the actual operation process, it is found that there are still many problems, such as the object in some objects It is difficult to obtain the ideal segmentation accuracy under the influence of the factors such as volume overlapping occlusion, light intensity, background interference and so on. This algorithm is based on the deep convolution neural network, and is improved on the basis of deeplabv3 plus network model. In order to slow down the grid effect and reduce the impact of the loss of spatial structure information on the semantic segmentation accuracy, the spatial void pyramid pooling module based on the mutual quality factor is used to improve the segmentation accuracy.<br>
正在翻译中..