Hyperspectral image camouflage target detection method based on deep learning

Abstract

The invention discloses a hyperspectral image camouflage target detection method based on deep learning, and the method comprises the following steps: a, constructing a hyperspectral data set: collecting, preprocessing, dividing and marking the hyperspectral data set, and obtaining a training data set and a test data set; b, constructing a target detection model: using an open-source Mask R-CNN model, adjusting the Mask R-CNN model for an input hyperspectral data set, and constructing the target detection model; c, model training: training the target detection model constructed in the step b by using a training data set; and d, testing the model: detecting and identifying the test data set by using the target detection model trained in the step c. According to the invention, the target detection model based on deep learning is used, and the spectral dimension and spatial dimension information of the hyperspectral image is used at the same time, thereby achieving the positioning and classification of camouflage targets.

Description

technical field [0001] The invention relates to the technical field of camouflaged target detection, in particular to a deep learning-based hyperspectral image camouflaged target detection method. Background technique [0002] Camouflaged targets refer to targets that are concealed using engineering and technical measures and using terrain and features. According to the movement characteristics of the target, it can be divided into fixed camouflage targets (electronic equipment, military facilities, etc.) and active camouflage targets (such as personnel, vehicles, ships, etc.). For fixed camouflaged targets, the traditional target detection method first searches the possible position of the target in the image through a sliding window, then uses SIFT, HOG and other features to extract the features of the target in the image, and finally inputs the extracted target features into SVM, etc. The classifier performs classification recognition. For moving camouflage targets, som...

AI Technical Summary

Problems solved by technology

Due to the large amount of data in a single image, hyperspectral images are mostly used for classification tasks, and are rarely used in target detection tasks, especially camouflaged target detection tasks.

In addition, target detection algorithms based on hyperspectral images are mostly used to detect abnormal targets and cannot identify specific types of targets.

However, the traditional hyperspectral image target detection algorithm is designed based on prior knowledge, so that the information in the hyperspectral image is often not fully exploited.

Images