Research on Image Segmentation Algorithm Based on Sparse Representation and Multi-Task Learning

Abstract

The results of image segmentation greatly affect the accuracy and correctness of image recognition. In this paper, image segmentation is taken as the research focus, combing and introducing the principles of multi-task learning, image sparse representation and tracking algorithm. An image segmentation model based on sparse representation and multi-task learning is designed, and an attention learning network AL-Net based on multi-task earning is proposed. The loss function of multi-task makes the contour of the segmented image smoother. The image segmentation model constructed in this paper is applied to the image segmentation of cucumber disease leaves, when the number of images in the training sample rises from 80 to 120, the recognition accuracy of the model in this paper stably stays at 90%, and the average recognition rate of seven diseases, such as leaf spot and scab, is as high as 90.49%, and the average time consumed is only 7.68s, which makes the effectiveness remarkable.