A Computer Vision System for Mushroom Detection and Maturity Estimation using Depth Images

Mushroom farms in the United States have grappled with a critical labor shortage, particularly in harvesting white button mushrooms (Agaricus Bisporus), which are widely consumed in North American diets. Mushroom production is labor-intensive processes that require manual monitoring by skilled workers due to the need for selective harvesting. This highlights the need for automation to ensure consistent and efficient harvesting while enhancing shelf life. Maturity assessment, primarily based on size and cap shapes, guides harvesting decisions by workers. The cap diameter is a key indicator, with mature mushrooms exhibiting expanded, flatter caps. Automating the identification of mature crops can optimize this process, reducing labor dependency and increasing productivity. The overall goal of this study is to automate mushroom detection and maturity evaluation, thereby optimizing and enhancing the efficiency of the harvesting process and increasing the mushrooms' shelf life by ensuring they are harvested at the optimal time. The specific objectives of this study are to (1) develop a novel image processing method for the detection and segmentation of mushrooms using depth and RGB images, (2) implement YOLOv8 for the segmentation of mushrooms to compare its performance with the proposed method, and (3) efficiently assess mushroom maturity using depth images. The proposed algorithm showed an F1-score of 0.92 for mushroom detection, compared to an F1-score of 0.89 for the YOLOv8 model using the same image dataset. Furthermore, the mushrooms are classified based on maturity using depth images to extract critical information about their diameter and slope. This extracted data is fed into a Support Vector Machine (SVM) to categorize the mushrooms as mature or immature. The classification accuracy achieved by SVM is 95.31± 4.2%. The initial results are promising, and in the future, the aim is to improve the segmentation process by refining the boundary identification in mushrooms using depth images and integrating machine learning and image processing methods for better results. The future work will also classify the mushrooms into different stages of intermediate maturity.