Study On High-Throughput Detection Method Of Soybean Flower Bud Development Based On Machine Vision

Due to the current research on the growth morphology of soybean flower buds,contact measurement technology is commonly used.Because it will damage plants in the detection process,and it is impossible to detect the continuity of the same flower bud object.Therefore,to solve this problem,nondestructive testing technology based on machine vision has important practical significance.In this paper,a set of machine vision system is constructed,which can realize online and nondestructive detection of the growth morphology changes of flower buds and study the morphological and structural changes of flower buds throughout the flowering period.The main contents and conclusions of the paper are as follows:(1)Through the analysis of the working requirements of the flower bud image acquisition system,the non-destructive testing system of plant flower bud image is constructed,which realizes the automatic,long-term,continuous and non-destructive acquisition of plant flower bud image,and completes the hardware selection and software design of the image acquisition system.(2)Through the analysis of the collected images of flower buds of plants,various image enhancement algorithms are compared and studied,and the median filtering method is determined to preprocess and analyze the images.Then the template matching algorithm based on gray scale is analyzed and the normalized cross correlation method is optimized for template matching of flower bud images.Finally,two experiments are carried out to optimize the best template size.The test showed that the matching degree of the template with size of 220*230 reached above 0.99,and the duration of matching and identifying flower buds was the longest,up to 132 hours.Therefore,the template with the size of 220*230 is selected as the best template for matching flower bud images.(3)In order to achieve the extraction of flower bud features,a variety of threshold segmentation algorithms were compared and studied,and the OTSU algorithm of adaptive threshold segmentation method was optimized.The morphological characteristic parameters of flower bud were calculated by shape transformation,regional difference set and speckle removal.Finally,experiment three is carried out to verify the accuracy of the detection algorithm for flower bud in this paper.The experiment compares the detection algorithm in this paper with the manual actual measurement method.The results showed that the correlation coefficients of the area and circumference of flower buds measured by the detection algorithm and the actual artificial measurement reached 0.9611 and 0.8983 respectively,and the average error ratio of the area was 5.713% and the average error ratio of the circumference was 4.716%.The error is relatively low,which indicates that the accuracy of this detection algorithm is high.