Research And Implementation Of Camouflage Personnel Detection Method Based On Multispectral Image

In modern warfare,camouflage camouflage and battlefield reconnaissance are indispensable tactical components.With the continuous improvement of camouflage clothing,it is becoming more and more similar to the surrounding vegetation environment,making it difficult for camouflage personnel to be accurately detected in visible light.This poses a great threat to battlefield reconnaissance.Multispectral imaging technology can simultaneously acquire the spectral information and image information of the target in the scene.Compared with visible light images,it has the characteristics of multiple bands and strong spectral resolution.Therefore,through the multispectral image of camouflage clothing,this paper analyzes the differences in spectral characteristics between camouflage clothing and the surrounding jungle environment,and carries out research on detection methods of camouflage personnel based on multispectral images.The main work and innovations of this paper include the following several aspects:(1)Since there is no publicly available multi-spectrum camouflage data set on the Internet,a multi-spectrum image acquisition scheme for camouflage staff is designed,a multi-spectrum image acquisition device is built,and a multi-spectrum data set for camouflage staff in a jungle environment is collected and constructed.(2)Feature band selection of camouflage clothing.According to the brightness information difference between the camouflage target and the jungle background,a band selection method based on image brightness information is used to select the characteristic bands of camouflage clothing,and the K-Means classification algorithm is used to verify the effectiveness of the feature bands selected by the method.Sex.The necessity of selecting a region of interest(ROI)during data processing is analyzed,and a reference whiteboard method is used to perform a reflectance reconstruction experiment on the original data in the region of interest(ROI).(3)A multispectral camouflage person detection method based on YOLOv3 network and image fusion algorithm is proposed and implemented.The NSST transform is performed on the characteristic band,in which the low-frequency components are weighted and fused using the principle of adaptive fuzzy logic,and the high-frequency components are fused using the large value of the modulus.The experiment proves that the camouflage target in the image obtained by the fusion algorithm proposed in this paper is outstanding,and all evaluation indexes are better than the traditional fusion algorithm.For fused images,YOLOv3 was used for target detection,and compared with the detection results of single-channel images,visible light images,and any combination of three-channel characteristic band images,it showed the advantages of multispectral images in camouflage target detection.(4)A multispectral camouflage target detection method based on semantic segmentation and U-Net network is proposed and implemented to solve the problems of too small camouflage targets and local occlusion.The correlation between pixels in multi-spectral images is analyzed.Based on the idea of semantic segmentation,pixel classification for multi-spectral images has greatly improved the classification accuracy compared with the visible light image segmentation results,and the camouflage target detection effect is significant..Based on the U-Net network,improvements have been made.By adding operations such as hole convolution layers and deepening the network structure,the segmentation accuracy is further improved,and excellent detection results have been achieved.(5)Based on the research content of this paper,a camouflage personnel detection system based on multispectral images is constructed,which can more intuitively and conveniently show the research results of this paper,and introduces the main functional modules and related implementations in detail.