An enhanced full convolution instance semantic segmentation algorithm suitable for small target detection

Abstract

The invention belongs to the technical field of image processing, and discloses an enhanced full convolution instance semantic segmentation algorithm suitable for small target detection based on a full convolution instance semantic segmentation (FCIS) algorithm, which comprises the steps of shared feature map extraction, preselection frame extraction, generation of a position sensitive score map,classification and regression. In the extraction process of the shared feature map, a conv1 feature map, a conv3 feature map and a conv5 feature map are fused, so that high-semantic information and high-detail information are reserved in the shared feature map; In the pre-selection frame extraction process, a dual RPN algorithm is provided for the poor network extraction effect of the pre-selection frame, and the average recall rate of the algorithm is increased by 7% compared with that of the RPN algorithm. The mAP of the EFCIS algorithm is improved by 3.5% compared with the FCIS algorithm, and for a small-size target, the mAP of the EFCIS algorithm is improved by 2.9% compared with the FCIS algorithm. Experiments show that the small target grabbing capacity can be improved very easily.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to an enhanced full convolution instance semantic segmentation algorithm suitable for small target detection. Background technique [0002] At present, the existing technologies commonly used in the industry are as follows: [0003] Scene understanding is a core difficulty in the field of computer vision, and instance semantic segmentation is a necessary process to achieve scene understanding. In the image field, instance semantic segmentation is a comprehensive task of collection image classification, target detection, and image segmentation. It is widely used In geographic information systems, unmanned driving, medical image analysis, robotics and other fields. [0004] With the rapid development of deep learning based on convolutional neural networks, more and more instance semantic segmentation subtasks can be completed using convolutional neural networks, ...

AI Technical Summary

Problems solved by technology

Therefore, in response to these problems, the present invention proposes an enhanced full convolution instance semantic segmentation (EFCIS) algorithm to solve the problem of poor performance of the full convolution instance semantic segmentation (FCIS) algorithm for small target segmentation

Images