Research On Image Fusion Of Infrared And Visible Image For Tree Identification

The harvester has promoted efficiency. However, because of the limit caused by single guidance, its capacity is seriously influenced. Thus, it is of great value to fuse multi-source images which contribute to provide reliably information for the harvester.Image fusion can not only reduce the pressure of data storage, but also prompt image’s quality and the amount of contained information. Moreover, on one side the fused image captures the detail and texture information from visible image, and on the other side it contains the contour effect form infrared image. Thus, while conserving information of both infrared and visible images, it becomes easier to extract information from fused image than form a single image.In this paper, image fusion algorithm on both pixel and feature level were proposed according to the characteristic of images of forest region. And the results of this study are as follows:1. A novel pixel level algorithm based on Contourlet transform and PCNN is proposed, which solve the block effect and distortion of gray level which is caused by the huge difference on pixel value between infrared and visible images. Then, compared with the mentioned algorithm, the result of the proposed method performs better on the quality assessment criteria.2. Based on the result of segmentation, a novel region-based algorithm is proposed, which could not only insure the quality of fused image, but also do good to enhance the detail information on target region. Thus, the accuracy of the data extracted from fused image could be reliable. In addition, the result shows that the main character in this study—energy of texture is promoted by 27.4%.3. With TLBO algorithm, the fused factor is optimized. Moreover, adjusting the range of the random parameter, the improved model finally contributes to a best fused result. The quantity of information maintained in fused image is promoted by 2.05%, while the Space activity is promoted by 15.27%.4. The recognition accuracy of pixel and feature level fusion result is 91.2%and 93.6%, promoted by 11.5%and 13.9%, compared with the original infrared samples, while by 16.4%and 18.8%, compared with the original visual samples. Based on this research, the conclusion can be draw, that pixel and feature level fusion have its own advantage and application scope. Pixel level method could prompt images’quality mostly. However, the time consumption and quantity of processed information affect its utilization. While, feature level method could highlight target, reduce time consumption, which makes it appropriate in forestry intelligent information detection.