Study Point Feature Matching Algorithm And Its Application In Stored Grain Pest Species Identification

As an important part of automatic navigation, medical diagnosis, pattern recognition andcomputer vision, image registration technique is playing a vital role in accomplishing theimage correction, change detection and image fusion. After years of research anddevelopment, image registration technique has made great achievements, especially playingan increasingly important role in pattern recognition, remote sensing, motion analysis andother fields. Image registration technique has continued to pursue the goal that the algorithmitself has robustness, precision, automation and high-speed registration.Stored grain insect detection is to monitor,forecast, early warn pests in stored grainand get its occurrence and dynamic development, using insect chemical ecology, moderncomputer pattern recognition techniques, image acquisition, transmission and compressiontechnique. The aim is to grasp the type, density, distribution, damage of pests and otherconditions, according to the information combined with the biological and ecologicalcharacteristics of pests along with environmental conditions, and forecast the developmenttrend of insects, and then provide scientific basis for prevention and control. In this thesis,image registration technique has been researched in a deep-going way, then the thesisattempts to apply SIFT (Scale Invariant Feature Transform) algorithm belonged to imageregistration algorithm for point feature and the improved algorithm to the stored grain insectdetection field, and the effect is obvious.Firstly, the thesis analyzes principle of HARRIS, SUSAN, and SIFTS three kinds ofpoint feature extraction algorithm deeply, and compares with the homologous experiment.Based on these studies, the thesis also optimizes and improves computing efficiency of theSIFT algorithm, using the improved SIFT operator for stored grain insect detection,implementing the feasibility and result analysis of SIFT algorithm in this field. The resultswell support that the SIFT feature extraction operator can be applied in this field, and the lastpart of the thesis defines a framework for a complete online stored grain insect identificationsystem.