Image segmentation method for enhancing edge and detail information by utilizing feature fusion

Abstract

The invention provides an image segmentation method for enhancing edge and detail information by utilizing feature fusion, and relates to the technical field of computer vision. The method comprises the steps of extracting features of an input image by using a convolutional neural network; inputting the extracted features into a decoding structure added with more feature fusion, and enriching theedge and detail information while recovering the image resolution to obtain a dense feature map; outputting maximum values of different classifications through a normalization method; and calculatinga cross entropy loss function, and updating the weight in the network by using a random gradient descent method. According to the method, the position and boundary detail information lost in the encoding stage can be recovered while the resolution of the feature map is recovered, the information of the picture is enriched, the dense feature map is obtained, the sparse feature map brought by directup-sampling is compensated, the segmented boundary and details are clearer, and the segmentation effect on detail small objects is improved.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to an image segmentation method for enhancing edge and detail information by using feature fusion. Background technique [0002] With the continuous advancement of science and technology and the rapid development of the national economy, artificial intelligence has gradually entered people's field of vision, playing an increasingly important role in human production and life, and artificial intelligence has been widely used in various fields , image semantic segmentation is an important research direction of artificial intelligence and a very important means to realize automatic scene understanding, which can be applied in many fields such as automatic driving system and unmanned applications. [0003] Image semantic segmentation technology is an important branch of computer vision in machine learning. Image semantic segmentation is to process the input image, automatically ...

AI Technical Summary

Problems solved by technology

[0004] After applying deep learning to semantic segmentation, many classic segmentation methods have emerged, such as the full convolutional network FCN, the SegNet network with an encoder-decoder structure, and the DeepLab with hole convolution. The pooling and downsampling of the image will lose the position information and boundary detail information of the picture, and this process is irreversible, and the removed information cannot be completely restored, so the feature map after upsampling in the decoding stage will become sparse due to the loss of information , these methods have certain limitations

[0005] The fully convolutional network FCN and the traditional SegNet network lose position and edge details due to downsampling, and the lost information is not reproduced when the upsampling is performed in the decoding stage, and the obtained feature map is sparse. Although the SegNet network passes through the pool The index restores the position information, and uses the convolution operation to enrich the boundary and detail information, but there are still a lot of missing information.

[0006] Atrous convolution is a convolutional layer that can obtain dense feature maps, but the computational cost of using atrous convolution is relatively high, and processing a large number of high-resolution feature maps will take up a lot of memory

[0007] The common problem of current image semantic segmentation methods is that the preservation of edge detail features and position information still needs to be further improved, and the accuracy of segmentation also needs to be improved.

Images