Seed Germination Detection Method Based on Lightweight YOLOv5

Seed germination test is an important task for seed researchers to measure seed quality and performance. In the seed germination test, a large amount of manual effort is usually required to collect data on seed germination and growth, which is a tedious, time-consuming, and error-prone process. Classic image analysis methods are not well-suited for large-scale germination tests as they often rely on manually adjusting color-based thresholds. Here, we propose an improved lightweight model called MB-YOLOv5 based on YOLOv5 for seed germination detection, which enables automatic detection of seed germination and significantly reduces the manpower and time costs of seed germination tests. The results show that the MB-YOLOv5 model achieves average accuracy rates of 99.3%, 99.1%, and 99.2% for germination detection of Zea Mays,Secale Cereale, and Pennisetum Glaucum seeds, respectively. Moreover, the MB-YOLOv5 model reduces the model size and floating-point operations by 77% and 85.4%, respectively, compared to YOLOv5s. This method provides a reference for the automation of seed germination experiments.