Automated Classification of High-Grade Dried Shiitake Mushrooms Using Machine Learning

This study aims to automate shiitake mushroom sorting using an anomaly detection system with Autoencoders (AE) trained on acceptable product data.Initial experiments using CNN approaches highlighted challenges in achieving high accuracy for acceptable product classification, necessitating improvement.The AE-based approach showed progress in detecting defective products via data cleansing, augmentation, and training optimization.However, misclassification of acceptable products with features like darker areas or complex textures remains an issue.This presentation outlines current findings and strategies, including data expansion and model improvements, to address these challenges.