Abstract:

In the standard germination test of seed, manual timing is required to collect data on seed germination and growth, such as germination rate, germination vigor, shoot length, and root length. However, this measurement process is time-consuming, laborious, and can easily damage germinating seedlings. To address these issues, a automatic detection method for maize seed germination is designed based on the Mask RCNN (region-based convolutional neural network) model and machine vision technology. First, within the 7-day germination test period of maize seeds, images for model training and testing are collected daily, and the seed positions are annotated using the Labelme tool, then a seed localization model is trained based on the annotated images. Second, based on the seed mask regions located by the model, an elliptical region for monitoring seed germination is defined, and the seed germination status is automatically identified. Finally, the main skeleton line of germinating seedlings is extracted using skeleton extraction and depth-first search algorithms, shoot length and root length are measured separately by calculating the centroid coordinates of seed masks. The results show that this method can effectively recognize germinating seeds and automatically measure indicators such as germination rate, germination vigor, shoot length and root length in maize seed germination experiments, providing a technical reference for the automation management of seed germination tests.

Key words: standard germination test, germination rate, mask region-based convolutional neural network, corn, skeleton extraction, shoot length, root length