An accurate machine learning approach for seed germination prediction

Abstract

To determine the quality of seeds, researchers often must manually check for seeds germination. The process is cumbersome, time consuming and error-prone since it requires the researcher to manually examine at least a few hundred to thousands of seeds. Hence, an automatic
seeds germination prediction solution is required. Over the years, with the help of deep learning methods, some studies have accurately
predicted the performance of seeds given just a picture of them. However, one downside of the deep learning approaches is the result does
not give more insight into which factors of the seeds’ image contribute to a successful germination process. In this paper, we propose a
classical machine learning method with a carefully designed features engineering process to both accurately predict seeds germination and
give more insight into the relevant factors for a seed’s germination process. At 95% prediction precision, the proposed method suggests that
relevant factors are: seed’s size, the circularity, brightness distribution and its skewness and kurtosis.