Deep learning-based robust analysis of laser bio-speckle data for detection of fungal-infected soybean seeds

Seed-borne diseases play a crucial role in affecting the overall quality of seeds, efficient disease management, and crop productivity agriculture. Detection seed-borne using machine learning (ML) deep (DL) can automate process at large-scale industrial applications for providing healthy high-quality seeds. ML-based methods are accurate detecting classifying fungal infection seeds; however, their performance degrades presence noise. In this work, we propose laser bio-speckle based DL framework detection classification seeds under varying experimental parameters noises. We develop DL-based spatio-temporal analysis technique data networks, including neural networks (NN), convolutional (CNN) with long-short-term memory (LSTM), three-dimensional (3D CNN), LSTM (ConvLSTM). The robustness models to noise is key aspect analysis. study, find that ConvLSTM model has an accuracy 97.72% on test robust different types noises 97.72%, 94.31%, 98.86%, 96.59%. Furthermore, (ConvLSTM) evaluated variations such as frame rate, size, number frames used. This also sensitive towards activity order, it shows average 99% four classes.