Mosaic of crop remote sensing images from UAV based on improved SIFT algorithm
Jia Yinjiang,Xu Zhenan,Su Zhongbin,Jin Siyu,Arshad M. Rizwan
DOI: https://doi.org/10.11975/j.issn.1002-6819.2017.10.016
2017-01-01
Abstract:The technology of unmanned aerial vehicle (UAV) remote sensing takes the UAV as a remote sensing platform and the digital equipment as the mission payload. UAV is characterized by high mobility, low cost and automated acquisition of the remote sensing data. These characteristics of UAV can be used to acquire high resolution images, and therefore, they are widely used in precision agriculture, pest forecasting, crop yield prediction, and so on. However, due to the restrictions of acquisition equipment and flight height, the task area may not be fully covered in acquired images. So it is necessary to mosaic the gathered image. In the process of using the scale-invariant feature transform (SIFT) algorithm for the mosaic of crop remote sensing images, the feature points were few and the effect of mosaic was poor because of the low contrast of the crop remote sensing images. To solve this problem, the contrast experiment was conducted for analyzing the steps of SIFT algorithm using the low altitude crop remote sensing images of past years. It was noted that the images were blurred because of Gaussian transform in the stage of image preprocessing and Gaussian pyramid constructing, which led to the lack of the feature points. Aiming at this problem, we proposed to use a sharpening filter to highlight the details of the image and increase the number of feature points after the Gauss transform. In the stage of searching feature points, after analyzing the feature points on the same scale, we found the feature points concentrated in the regions where the color changed obviously; while in the different scales, the overall distribution of the feature points was inversely proportional to the scale. To prevent the reduction of the feature points caused by the same sampling step in the regions of high scale, in this paper, a search method of feature points based on adaptive sampling step was used to achieve uniform distribution of the feature points, because the uniform distribution of feature points was more conducive to image mosaic. A larger sampling step was used in the low-scale region and a smaller sampling step was used in the high-scale region. In the phase of feature matching, the feature points of crop remote sensing images were mostly concentrated at the minimum value, and a few maximum points often did not affect the image matching effect, but it would increase the amount of calculation. Therefore, this paper adopted non-minimum suppression to improve the efficiency of the algorithm. In order to verify the validity of the proposed algorithm, we carried out the experiment based on UAV remote sensing platform from July to August in 2016. Through the contrast experiment using SIFT algorithm and proposed algorithm, we could see that, as for low altitude crop remote sensing images, the number of feature points was averagely increased by 77.5% after optimization, the distribution of feature points was also more uniform, the number of matching points averagely increased by 15 pairs and the matching rate averagely increased by 9.46%; for the crop images that could not be stitched by SIFT algorithm, the proposed algorithm could extract more than 8 pairs of matching points, and this satisfied the condition of image mosaic. Experiments have showed that the custom sharpening filter can extract more feature points, and after introducing the adaptive sampling step, the distribution of the feature points is more uniform, and the optimized algorithm keeps the robustness. Therefore, this algorithm is more suitable for the crop remote sensing images mosaic than SIFT algorithm.