FIGURE 7.

Network architecture of the HairNet deep learning model to score cotton leaf hairiness. HairNet consists of four main parts. First a leaf surface images are passed through a Data Augmentation module (A) that augments each image by applying a variety of image processing techniques. Processed images are then passed to a Feature Extraction Network (B) that extracts discriminative visual features from the image representation. Extracted visual features are then passed to a simple Classification Neural Network (C) that assigns each input image to a specific leaf hairiness score. Raw scores are then processed by the Leaf Hairiness Scoring module (D) which generates three accuracy metrics for scoring cotton leaf hairiness. Adapted from Rolland et al. (2022) and is licensed under CC BY 4.0.