Research On Automatic Crop Seed Counting Method Based On Image Recognition

Efficient and accurate automatic counting method for crop seeds is beneficial to promote the development of related work in modern agricultural production,such as calculating1000-seed weight and breeding,etc.,without seed counting.Among them,1000-seed weight is an important indicator for measuring seed quality and predicting yield;breeding innovations promote the transformation of agricultural development methods,and have important significance in terms of food security and ecological security.The widely used manual and photocell counting methods are cumbersome,costly,inefficient,and accurate.With the development of information technology,in order to meet the rapid and accurate counting environment for a large number of seeds in modern agricultural production,an automatic crop seed counting method based on image recognition is proposed.To achieve seed counting based on image recognition,the first step is to collect seed images.Then the seed image is pre-processed: basic image pre-processing operations including graying and filtering;initial segmentation of the image by combining edge detection and threshold binarization;morphological processing of the initial segmentation of image;and labeling of connected areas.After the pre-processing,the seeds are effectively segmented from the background of the image,and most of the weakly adhered seeds are also separated,but there are still some deep-adhered seed regions.If you directly count them,the error will be greater.This article analyzes two existing methods for segmentation and counting of adhesion seed images.For the case of poorly segmented adhesion seed regions,an image recognition-based method is used to identify the type of seed adhesion and then accurately segment it to achieve accurate seed counting: including Set the circularity threshold to determine the adhesion seed area;model the seed adhesion type and extract effective classification features as input parameters of the SVM;use the SVM classifier to identify the seed adhesion type;establish segmentation rules for different seed adhesion types to achieve accurate Adhesive segmentation;the number of connected domains is detected and the seed count is completed.The verification of the counting method in this paper mainly uses soybean as the experimental object,and counts a random number of soybean seed images.The experiments show that the article counting method is fast and the counting accuracy is high.This counting method still has certain applicability when applied to other crop seed images,and has practical significance in obtaining 1000-seed weight of seeds,breeding and other agricultural production-related work.