Research On Multi-sensor Image Fusion Technology Based On Machine Learning

With the development of image processing technology,image fusion technology has gradually become an important part of it,and the application range of image fusion technology is increasingly wide.Ocean exploration is one of the application directions.The ocean environment is relatively complex,and ocean monitoring will be affected by time,weather and other factors.The imaging principles of polarization sensor,visible light sensor and infrared sensor are different,and they have their own advantages and disadvantages.Polarization sensor has the ability of "penetrating cloud and fog",and it can also overcome the influence of sea flare on the target,but the image information richness is poor;infrared sensor mainly detects the target according to the thermal radiation difference between the detected target and the surrounding environment,and the brightness and weather environment will not affect the infrared imaging basically;visible light sensor has rich information and is easily affected by the environment.Therefore,the fusion of the three can learn from each other and achieve the goal of target detection.However,for infrared,polarized and visible image fusion,how to improve the fusion effect and reduce the amount of calculation is relatively.In order to solve the above problems,this paper adopts the multi-sensor fusion algorithm based on machine learning.The overall idea of the algorithm is: the image input system obtained by the pre-processing polarization,near-infrared and industrial visible light sensors.Firstly,we use the improved CNN(Convolutional Neural Network)model to extract the feature data of polarization image and visible image,and then use NSCT(Non-Subsampled Contourlet Transform)to decompose the feature fusion image and infrared image.Finally,the weighted average fusion rule is used to reduce the calculation amount for the low-frequency subband image,and the pixel maximum fusion rule is used for the high-frequency subband image to improve the edge retention and get the final fusion image.Compared with NSCT and WT(Wavelet Transform)fusion algorithm,the structure similarity,average gradient and spatial frequency of the fused image are improved by 4.4%,38.9%,37.2% and 17.5%,46.3% and 27.8% respectively.It makes full use of the infrared image brightness,intensity information and polarization image to improve the image quality,so that the target highlights the background clearly,so as to achieve the purpose of clear recognition of sea targets.