Research On Camouflage Recognition Based On Hyperspectral Imaging

Hyperspectral imaging technology has developed rapidly in recent years,and has been widely used in water,agricultural soils,food safety,military and other fields because of its high spectral resolution,wider band,and image and spectral fusion techniques,which brings great opportunities for military camouflage reconnaissance.With the increasing application of high-tech in the modern battlefield,camouflage can improve the survival rate of weapons and personnel in the battlefield,which makes military camouflage become the focus of research.This paper is based on the principle that differences in the characteristics of different types of battlefield backgrounds and camouflage materials are directly mapped onto differences in spectral characteristics.The spectral and spatial information of the simulated battlefield background and its camouflage materials is obtained through hyperspectral imaging technology and combined with mathematical statistical classification models to achieve the classification and identification of the mapping data.Reflective grating and push-scan hyperspectral imaging system built in a laboratory optical darkroom environment,and the hyperspectral image information of the simulated desert background,simulated jungle background,desert camouflage net,jungle camouflage net,jungle camouflage clothing and desert camouflage clothing samples in the wavelength range of 400.00 to 1000.99nm（visible light,near infrared）was acquired with an average spectral resolution of 2.5nm.The region of interest extraction of hyperspectral camouflage sample images is achieved by pre-processing such as black and white correction,multiple scattering correction,and image segmentation to remove background.The relevant principal components,loading curves and the reference spectral curves under the corresponding feature space are obtained after dimensionality reduction of the data of the six categories of samples to be analyzed by the principal component analysis algorithm.After averaging the image elements in the region of interest and their nearby neighborhoods,58% are used as the training samples and the remaining 42% as the test samples,and the classification models established by the random forest algorithm,support vector machine algorithm and convolutional neural network algorithm are designed and optimized using their similarities and differences with the reference spectra.The results show that the recognition rate of the convolutional neural network algorithm is over 99%,which is the optimal classification model.Comparing the four camouflage methods,the recognition rate of desert camouflage suit is the best;comparing the two camouflage materials,the recognition rate of camouflage suit is higher than that of camouflage net.This study verifies the scientific practicality of hyperspectral imaging technology for camouflage identification classification on the modern battlefield,which has certain guiding significance for camouflage reconnaissance in the future intelligent battlefield.