Sugarcane image stitching under transverse transport based on improved SURF algorithm

This paper proposes a sugarcane image stitching algorithm based on an improved SURF method to capture high-quality, wide-field images of complete sugarcane stalks. To enhance registration accuracy, artificial markers are introduced into the background, helping to address the challenges posed by the smooth surface of sugarcane and low feature point matching precision. Additionally, a mesh segmentation technique combined with an enhanced SURF algorithm is used for feature extraction, which tackles issues such as uneven feature distribution and slow processing speed caused by global image feature extraction. A double screening registration method is also proposed to further improve the accuracy of image mosaicing. To reduce stitching gaps, an image fusion technique based on the optimal suture line is employed. Experimental results show that the algorithm has an average runtime of about 2900 ms, slightly longer than the ORB algorithm at 2000 ms but significantly faster than the original SURF at 4200 ms. In terms of stitching quality, the average image information entropy is 6.34, which is higher than both the SURF (6.325) and ORB (6.075) algorithms, indicating better image quality.