Soybean seed counting and broken seed recognition based on image sequence of falling seeds

Seed counting and broken seed identification are important tasks in evaluating seed quality. In this study, we proposed a computational method designed to perform these two functions. Image sequences of soybean seeds during falling were collected, and their morphologies were examined from different views. An a priori clustering algorithm composed of a support vector machine and k-means clustering algorithm was used to segment touching seeds within images. The morphologies of specific soybean seeds in sequential images were associated based on a forced neighbor association criterion to avoid repeated counting and obtain shape features from multiple views. Based on the areas in different views, the basic shape features in the initial multi-view shape features were sorted to obtain the guided multi-view shape features. The support vector machine was used with the guided multi-view shape feature to classify seeds as intact or broken. The experimental results show that the proposed a priori clustering algorithm accurately segmented touching seeds. The forced nearest-neighbor data association algorithm is insensitive to touching seeds and achieved highly accurate seed counting. Compared with the single-view shape feature, the multi-view shape feature significantly improves the accuracy of seed morphological classification. The proposed method exhibited considerable potential for applications in agricultural engineering.

Automated summary

This study proposed a computational method for seed counting and broken seed identification. It collected image sequences of soybean seeds and examined their morphologies from different views. The method accurately segmented touching seeds and improved the accuracy of seed morphological classification by using multi-view shape features. It showed great potential for applications in agricultural engineering.