Humanoid target segmentation method based on convolutional neural network

Abstract

The invention discloses a humanoid target segmentation method based on a convolutional neural network, belongs to the field of computer visual identification, and is used for solving the problems of low humanoid target segmentation precision, low humanoid target identification and segmentation speed and low adaptability to complex scenes at present. The method comprises the following steps: determining a target image data set of a human-shaped target, extracting image features through a deep convolutional network, and constructing an FPN network by using an obtained feature map; inputting the FPN network features into a classification network and a segmentation network at the same time, and performing the two networks in parallel; and fusing output results of the two networks to obtain a segmented human-shaped target feature map, and finally, intercepting a human-shaped target area from the image by determining position information of the human-shaped target, eliminating background interference, and improving subsequent bullet hole detection speed and precision. According to the method, rapid, efficient and high-adaptability humanoid target segmentation is realized, the humanoid target segmentation precision and speed are improved, and correct target surface data are provided for subsequent target shooting detection processing.

Description

technical field [0001] The invention belongs to the technical field of computer vision recognition, in particular to a humanoid target segmentation method based on a convolutional neural network. Background technique [0002] The current automatic target reporting system mostly adopts the principle of target reporting based on image processing, and the segmentation of humanoid targets is a very important step. The existing humanoid target segmentation technology mostly uses image-based color, texture and shape characteristics to complete through traditional image processing technology, which has low adaptability to complex scenes, and the accuracy of target surface segmentation needs to be improved. Moreover, the rough target surface segmentation will affect the judgment of whether the bullet holes fall in the effective area, resulting in statistical errors in target reporting. [0003] With the rapid development of deep learning convolutional neural network and the substan...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the problems of inaccurate humanoid target segmentation and slow humanoid target recognition speed in the prior art, the present invention proposes a humanoid target segmentation method based on convolutional neural network, which extracts image features through deep convolutional network, and uses the obtained The feature map of the FPN network is constructed; the FPN network features are input into a classification network and a segmentation network at the same time, and the two networks are executed in parallel; after the output results of the two networks are fused, the segmented humanoid target feature map is obtained, and finally the humanoid target is determined. The location information of the humanoid target area is intercepted from the image, the interference of the background is eliminated, and the speed and accuracy of subsequent bullet hole detection are improved.

Images