Non-Invasive Detection of Early Germination in Beans Using Laser Speckle Temporal Analysis and Deep Learning

In their early stages, germination of bean is not externally visible, requiring invasive methods such as physically breaking the seed for detection. This study proposes a non-invasive and automated approach that combines laser speckle imaging with deep learning to identify early-stage germinated defects in beans. Speckle images were captured under coherent laser illumination (λ= ۶۳۳ nm) for samples subjected to germination periods of ۰, ۶, ۱۲, and ۲۴ hours. Two contrast analysis techniques were evaluated: Laser Speckle Spatial Contrast Analysis (LASCA) and the more advanced Laser Speckle Temporal Contrast Analysis (LASTCA), the latter using temporal intensity fluctuations across ۱۲۰ video frames. From the resulting contrast maps, regions of interest (۵۰×۵۰ pixels) were extracted and used to train a Convolutional Neural Network (CNN) for binary classification. The proposed system achieved high performance, with an accuracy of ۹۲.۳۳% and sensitivity of ۹۸.۲۱%, successfully detecting germinated defects invisible to conventional inspection. By integrating temporal bio-speckle analysis with deep learning, this method offers a scalable solution for intelligent, non-destructive grain inspection—addressing a critical gap in current computer vision systems and contributing to the advancement of Agribusiness ۴.۰.

کلیدواژه ها:

Computer Vision ، Convolutional Neural Network ، Laser Speckle image