Deep Learning Algorithms in the Field of Computer Vision Art

Abstract

In order to solve the problem of poor recognition performance and long time for key targets in visual sensing images, the author proposes the research of deep learning algorithms in the field of computer vision art. The author uses color feature extraction to identify the feature quantity of the target, and constructs a convolutional neural network model with multiple feature parameters by calculating the texture feature differences between the visual sensing image target and the background area. Using the exponential Laplace loss function to reduce the amplitude of intra class feature changes in the model and adjust the distance between features of different class centers. Combining adaptive block labeling to complete the entire process of target enhancement recognition. Design testing experiments based on visual sensing images of flames and vehicle monitoring. The experimental results show that as the noise in the image gradually increases, this method has a good enhancement effect on visual sensing images, and the average gradient information of the enhanced visual sensing image is higher than 0.82, which is less affected by noise and has a good enhancement effect. The method proposed by the author has high practical application value.

Keywords: visual sensing image, Convolutional neural network, target recognition, differences in texture features, block marking