Research On Infrared Multi-band Intelligent Detection Technology Of Weak And Small Target In The Air

In modern military warfare.As the air battlefield environment becomes more and more complex,infrared weak-small target recognition detection under multi-band detectors has been a research hotspot and difficult problem in the field of infrared image processing.In order to improve the recognition ability of the detector,it is necessary to use dual-band imaging detection technology to obtain infrared images of objects in two different wavelength bands.At the same time,the fusion of radiation characterisics under different spectra makes it possible to accurately detect weak-small targets at long distances.Based on this.This paper uses dual-wave(3-5um and 8-14um)simultaneous detection to obtain dual-band image data for the problem of anti-background interference of "weak" and "small" targets in infrared images.Analyze and compare the characteristics of weak-small targets in dual-band images.Integrate images obtained from different sensors,and use certain preprocessing method and intelligent fusion algorithm to achieve complementary image difference information,while reducing complex background information and enhancing target brightness.Then,based on the intelligent detection algorithm,the weak-small targets are identified and detected,thereby reducing the false alarm rate and accurately obtaining the weak-small target information in complex dynamic scenes.The main research carried out in this paper is as follows:1.Aiming at the problems of dual-wave long-distance detection target with various characteristics,weak-small target imaging dispersion,and difficult to evaluate images.Firstly.Based on the detection radiation characteristics of the target under the detector,the imaging characteristics and radiation characteristics of the target under the dual-wave detector are studied to provide theoretical support for the difference complementation of the dual-band image fusion.Secondly.The target imaging dispersion model of infrared detection system effects is proposed,the simulation is verified by theoretical analysis and experimental testing.The model determines the dispersion size of weak targets after imaging,and provides technical support for removing the blur of weak target edges after image fusion.Finally.Aiming at the problems such as blurred edge and information distortion of infrared weak-small target image,the objective evaluation index of the image is proposed.2.Aiming at the problem that the traditional fusion algorithm is difficult to effectively enhance the brightness of weak-small targets under the undulating clouds in the air.A deep fusion algorithm of infrared weak-small target images against background radiation interference is proposed.The algorithm combines wavelet transform and depth feature extraction(DWT-FE)methods.Firstly.Discrete wavelet transform is performed on the image to obtain high-frequency and low-frequency components of the image,and then different fusion rules are used according to different components.At the same time.The quad-tree depth feature decomposition method is used to extract the detailed features of the dual-wave image.Finally.The reconstructed fusion image,the medium-wave detail feature image and the long-wave detail feature image are gray-scale modulation fusion to obtain the final fusion image.Using the proposed algorithm and a variety of classic algorithms to fuse different background dual-wave data.The experimental results show that the algorithm proposed in this chapter makes the aerial target and background texture clear,and the brightness of the weak and small targets in the image is enhanced,which has a significant effect on improving the image signal-to-noise ratio.Compared with the weighted fusion result,the average peak signal-to-noise ratio can be increased by13.3%.At the same time.It also shows that the proposed algorithm has an obvious effect on the fusion of the dual-wave image of the ground object background and the sea background.Thereby providing theoretical support for multi-scene image fusion in engineering applications.3.Aiming at the problem that the traditional fusion algorithm and deep learning fusion algorithm can not adaptively highlight the weak-small targets and suppress the radiation of bright and dark clouds.An intelligent image fusion algorithm for weaksmall targets against bright and dark cloud radiation is proposed.The algorithm combines intelligent latent low-rank representation and wavelet transform(Lat LRR-DWT)method.Firstly.Lat LRR is used to intelligently train all source images into L matrix for extracting salient features.At the same time.The original image is decomposed into high-frequency details and low-frequency contours through DWT.Then.The high frequency part adopts the fusion method of taking the absolute value to obtain the cloud details and target details.And the low frequency part adopts the weighted average method to obtain the background contour information.On this basis.The training matrix L and the high frequency fusion part are used for adaptive contrast modulation fusion to extract image detail information.Finally,the contour part and the characteristic part are combined to reconstruct the fusion image.The Lat LRR-DWT method and nine traditional/deep algorithms are used to fuse the air cloudy background data,which shows that the algorithm has obvious advantages in highlighting the characteristics of weak-small targets and suppressing the air cloudy background.At the same time.Compared with the weighted fusion result,the peak signal-to-noise ratio can be increased by 24.7 %.Thus verifying the practicability of the fusion algorithm to improve the brightness of weak and small targets,suppress the cloud background,and enhance the image signal-to-noise ratio.4.In view of the cloud background.The target is too small and the general convolutional neural network algorithm is difficult to effectively identify and detect.Therefore.An improved YOLOv3 weak target detection and recognition algorithm framework(YOLOv3-Vo VNet)based on Vo VNet is proposed.By up-sampling the8-fold down-sampling feature map output by YOLOv3,the 2-fold up-sampling feature map is stitched with the 4-fold down-sampling feature map output from the second residual block in Darknet53.The output feature map can be obtained.The pixel value is less than 4×4 corresponding to the weak target.So that the network can obtain more feature information of the weak-small target,and improve the detection rate of the weak-small target in the image.At the same time.The Vo VNet network lightweight model downsampling is used to increase the number of feature channels and quickly detect weak-small aims.The experimental test data uses long wave images,medium wave images,fusion images under the sky background in our own data set,and medium wave images under the sky background in the public data set.The comparison of the results of the YOLOv3-Vo VNet detection algorithm and Faster-RCNN,SSD,YOLOv3,FPN detection algorithm show that the proposed algorithm has improved recall and detection accuracy to varying degrees.Especially in the target detection accuracy of its own data set fusion image Reached 92.57%,and the target detection accuracy of wave images in the public data set can reach 93.65%.It shows that the proposed algorithm can resist the interference of sky and cloud background while ensuring high accuracy.In summary.This paper deeply studies the problem of infrared multi-band intelligent recognition of weak targets in the air and provides corresponding solutions.It achieves the good results of the algorithm in dual-band image fusion and weak target image recognition detection.The algorithm is provided for practical engineering applications.