Research Of Tobacco Pest Image Registration Based On Improved ORB Algorithm

The fast detection and recognition of pests in tobacco is the basis of tobacco pest and disease control in agriculture. Traditionally, experts and tobacco growers observe the external features of pests and then compare these with specimens while identifying pests, which is time-consuming and labor exhaustive. With the development of computer technology, image registration technique is widely used in many fields of society and economy, such as agriculture, medical, remote sensing, industrial, military, etc. Feature extraction and feature matching are the main research contents of image registration techniques based on image feature, and also the central issues of algorithm improvement. The feature extracted by SIFT, SURF, ORB algorithm is classic local invariant feature, and these three feature extraction algorithm have their advantages and disadvantages in many respects, such as scale, rotation, translation, illumination, affine change, computing speed and the number of feature points matched. Due to their good performance, image registration technique based on the SIFT, SURF and ORB algorithm is getting more and more focused, however, the study of tobacco diseases and pests image registration technique is less.Typical tobacco pests were studied as research objects in this thesis. Image segmentation, feature points extraction, feature point matching, etc. were studied based on tobacco pests images with the technologies of Image registration. Then comparing the three feature algorithm based on experimental results, improved ORB algorithm with SIFT algorithm and RANSAC. The main research contents of thesis are as follows:(1) Collected the standard and to be matched tobacco pests images, including helicoverpa assulta, smoke green stink bug, tobacco beetle, hawkmoth, beatle, phyllotreta striolata, black beetle, then segment these images used iterative threshold segmentation algorithm.(2) The three algorithm of extracting image feature points were studied. SIFT, SURF, and ORB feature algorithms were used to extract feature points of tobacco pests. Improved KD-tree nearest neighbor query algorithm was used to match feature points. RANSAC algorithm was used to excluding mismatched feature points.(3) Evaluate the performance of SIFT, SURF and ORB feature algorithms based on experiment results, then comparing the three feature algorithms in two respects, one is the number of matching feature points, and the other one is calculation speed. Improved ORB feature algorithm with SIFT feature algorithm which has the advantage of scale invariant feature transform. In order to eliminate more mismatches, thus optimize feature points samples to improve the RANSAC algorithm.