Research On Agricultural Pest Image Matching Technology Based On Improved ORB

Image matching technology is the basis of multiple research directions such as image recognition and image stitching.With the continuous development of agricultural informatization,image processing technology has been continuously applied to many aspects of agriculture.Among them,in the identification of agricultural pests,the use of image processing technology to identify the types of pests in a timely manner,replacing the traditional working mode of using human eye recognition on the spot based on experience,is of great significance for pest control and agricultural production.The application of image matching technology in agriculture has also become a hot research topic.This paper mainly studies the feature matching algorithm based on ORB image.In the acquisition of agricultural pest images,pictures in the same scene will inevitably undergo various adverse changes such as scale changes,rotation changes,and lighting changes.These changes will have an impact on image matching.The ORB algorithm has the advantages of small storage space,light weight,and high real-time performance.It has invariance to rotation,illumination,and translation,but does not have scale invariance.The algorithm also has uneven distribution of feature points and is sensitive to image noise.The disadvantage of poor algorithm robustness.Currently popular image feature matching algorithms such as SIFT,SURF,BRISK,etc.all have invariance to the scale,rotation,and illumination of the image,but they are much slower than the ORB algorithm in operation speed.This research is based on the background of image matching technology,introduces the progress and principles of related technologies in detail,discusses and compares the mainstream algorithms,selects an algorithm suitable for agricultural pest image matching,and addresses the defects of the algorithm Make improvements,complete the optimization of the algorithm,highlight the principles,steps,advantages and disadvantages of the ORB algorithm,and propose three improvements to the problems of the ORB algorithm:(1)Feature extraction step in image feature matching algorithm based on improved ORB: In the process,the scale information function of the feature points is added to the Harris corner response function,so that feature points have scale information,a feature descriptor with scale information can be generated,the algorithm has scale invariance,and sub-pixel interpolation is performed on the feature points.Precise positioning,while adding a Gaussian function to suppress noise in the function.(2)ORB image feature descriptor based on the fusion BRISK algorithm: In the step of generating the ORB feature descriptor,Introduce the idea of BRISK algorithm,perform uniform sampling mode on the feature points,construct concentric circles with different radiuses using the feature points as the center,obtain N equally spaced sampling points on each circle,and perform Gaussian filtering on each feature point.Calculate the local gradient,and finally generate a binary feature descriptor to eliminate the clustering phenomenon and the sensitivity to noise at the feature points of the original algorithm,which enhances the robustness of the algorithm while retaining the rotation invariance of the original algorithm.(3)Elimination of mismatch points based on improved RANSAC: Sort the data points in the data set and select The points with the best matching effect are selected and fitted first,so that the optimal parameters are obtained in advance;the model is assumed to be that the number of correct matching points in the set is greater than the number of outer points or noise points,and then the parameters in the model are one by one The verification has solved the problems of instability and complicated calculation process of the original algorithm in specific scenes.Finally,the improved RANSAC algorithm was applied to the improved BRRB image matching algorithm in this paper.After matching experiments on agricultural pest images,the correct matching rate of the improved algorithm in scale experiments is 72.54 percentage points higher than the original algorithm,which effectively solves the defect that the ORB does not have scale invariance and retains the original algorithm.High efficiency in calculation speed and invariance to rotation,lighting,and affine transformation.