A Counting Detection System For Crop Seeds Based On Machine Vision

The grain weight of crop seeds is a key indicators for measuring the quality of crop seeds,and There is no more critical and tedious task than to calculate the number of crop seeds accurately when calculating the grain weight of crop seeds. Nowadays,Artificial and photocell are two mainly way for counting number of the crop seeds in our country,and both of them have the shortcomings of high-cost and low-accuracy.For this phenomenon, A system for automatic counting crop seeds with a combination for computer vision,image processing and mode recognition is designed. The system firstly wipes out the noise smoothly of the input images of seeds and other necessary pre-processing operations,and then partitioning the seeds images with improved fuzzy C-means clustering algorithm(FCM),after the initial segmentation, seeds and background is separated clearly and the individual seeds can be calculated. However,there are still some sticking seeds deeply in the image, large errors must be happened if counting the seeds directly,and the accuracy will be very low. So we first build the modes for different adhesion types of the deeply sticking seeds,and then building a support vector machine(SVM) classifier by extracting the specific characteristics of different adhesion types of those deeply sticking seeds,and ultimately we can identify the different adhesion types of sticking seeds in input images,and then achieving the goals of segmentation and counting by the different rules to different adhesion types.Using the corn seeds as examples,the designed system does many experiments with seeds of different numbers and different configuration,and the results show that the system has a high precision and anti-interference and low time consuming, when counting the numbers of crop seeds,There are some practical significance and universal application value in the field of calculating the grain weight of seeds and the modernization of agricultural production.