Infrared small target image labeling method and system

Abstract

The invention relates to an infrared small target image labeling method and a system. The method comprises the following steps: converting an original image in video data acquired by infrared acquisition equipment into an 8-bit single-channel image; splicing the three frames of single-channel images with similar time into a three-channel image; marking each three-channel image by using a marking frame; and determining the centroid of the mark box according to the gray value of each point in the mark box, and readjusting the size and position of the mark box through the centroid. Three frames of single-channel images with similar time are spliced into a three-channel image, and time sequence information can be reserved in the single image; then, a target in the three-channel image is marked; and finally, the mass center of the mark box is calculated. The size and the position of the mark box are adjusted according to the mass center, the target and the background within the set range around the target are contained in the mark box, time domain information and richer space domain information are fused into image marks according to a small target neural network model, and the accuracyand the practical value of the model in infrared investigation monitoring equipment are improved.

Description

technical field [0001] The invention relates to the field of infrared image processing, in particular to a method and system for marking an infrared small target image. Background technique [0002] In the current big data era, as a data-driven technology, the deep convolutional neural network learns and trains the model with the support of massive labeled data, builds a high-performance model, and performs image processing such as target detection, recognition, and tracking. field has achieved great success. With the support of rich and diverse massive labeled data, the deep convolutional neural network model has a strong generalization ability, and can be adaptively adjusted as the data changes to adapt to different application scenarios. In addition, the excellent performance of the convolutional neural network also benefits from its large and complex deep structure, which enhances the nonlinear fitting ability of the model. The convolutional neural network has a weight...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the technical problems existing in the prior art, the present invention provides a small infrared target image labeling method and system to solve the problem that the training sample features obtained by the existing labeling method are not rich enough

Images