Image semantic segmentation method based on multi-feature fusion

Abstract

The invention discloses an image semantic segmentation method based on multi-feature fusion. The method comprises the following steps: firstly, constructing a cavity dense structure fusing enhanced features and image edge features; respectively inputting a to-be-segmented image into the image enhancement channel, the cavity convolution channel and the image edge feature extraction channel for feature extraction; combining the extracted features in a dense connection mode; transmitting the combined features to a plurality of dense blocks formed by convolution of three layers of holes, and finally obtaining a result after pixel classification through a deconvolution layer. According to the method, the smoothness of the L1 norm is improved by using a variance fitting method, and redundant convolution kernels existing in a convolution layer are cut off by using the improved L1 norm. According to the image semantic segmentation method, the image semantic segmentation effect is improved under the condition of moderate convolution layer parameters.

Description

technical field [0001] The invention relates to the technical fields of image processing and machine vision, in particular to an image semantic segmentation method based on fusion of multiple features. Background technique [0002] Image Semantic Segmentation (ISS, Image Semantic Segmentation) can obtain image information according to the characteristics of the image, and is one of the focuses of digital image processing research. At present, the general neural network model can show a good segmentation effect for objects with obvious category attributes, but for some objects with similar attribute categories or complex image backgrounds, the common image semantic segmentation method will generate more Features often result in unsatisfactory model segmentation due to problems such as resolution degradation and insufficient local feature extraction. At present, in many image semantic segmentation methods, dilated convolution layers (Dilated Convolution) are often used to con...

AI Technical Summary

Problems solved by technology

However, in complex image semantic segmentation scenarios, the dilated convolutional neural network is affected by the differences in light, angle, and state in the image, as well as the high similarity between different types of objects, making the dilated convolutional neural network change a variety of In the case of empty convolutional layer parameters, it does not improve the segmentation effect of image semantics very well.

[0003] In addition, in the process of designing a neural network, it is usually chosen to provide more parameters than required, so that the convolutional layer and the dilated convolutional layer require huge computing resources, and it is difficult to meet the requirements for computing performance in practical applications or computing platforms. , which requires the image semantic segmentation method to reduce redundant parameters in the neural network while maintaining a good segmentation effect

Images