Bullet hole recognition method based on deep learning

Abstract

The invention discloses a bullet hole recognition method based on deep learning. The method comprises the following steps of 1, building a model; 2, collecting data; 3, processing and labeling the data; 4, training the model; and 5, recognizing bullet holes. A recognition system used in the method has low requirements on a target surface; the probabilities of false detection, missing detection andduplicate detection are below 1%, the recognition accuracy of the bullet holes is high, and the use requirements are met; the detection delay of the recognition system used in the method is less than40ms, and a detection result is synchronously displayed on a real-time video stream; under the condition of dense bullet holes, the overlapping rate of the bullet holes is smaller than 50%, namely, the bullet holes and the overlapping bullet holes can be distinguished; when target paper shakes relative to a camera, a center point alignment algorithm is used, so that the relative positions of thebullet holes and the center are unchanged, and the influence of the shaking of the target paper on the detection is eliminated; and a fuzzy detection algorithm is used for filtering out fuzzy pictureframes during hit by bullets, thereby effectively solving the problem of false recognition caused by picture fuzziness.

Description

technical field [0001] The invention relates to the field of shooting training, and in particular provides a method for recognizing bullet holes based on deep learning. Background technique [0002] Live ammunition shooting is a basic training assessment item for public security, armed police, military and other departments. At present, the target reporting method of these departments mainly adopts manual target reporting. Make a target report. This method of target reporting affects the efficiency of shooting training, and has high requirements for the quality of the target reporting personnel, and cannot report the number of rings in real time. [0003] In the prior art, most bullet hole identification methods use traditional image processing methods to calculate the difference between two adjacent frames of images, use the edge detection algorithm to find the contour information of the picture after the frame difference, and calculate the absolute area confidence of each...

AI Technical Summary

Problems solved by technology

However, this method will cause many false detections and missed detections in the case of large virtual targets, blurred images, and continuous holes caused by large target position shaking.

Images