Technological Development of Automated Harvesting for Cultivated Button Mushroom Using Image Processing

• Published in: Periodica polytechnica. Electrical engineering and computer science Vol. 68; no. 4; pp. 413 - 423
• Main Authors: Hubay, Csongor, Geösel, András, Hubayné Horváth, Nóra, Géczy, Attila
• Format: Journal Article
• Language: English
• Published: Budapest Periodica Polytechnica, Budapest University of Technology and Economics 01.10.2024
• Subjects: Algorithms; Automation; Harvesting; Image processing; Image quality; Image segmentation; Mushrooms; Picking; Programming languages; Python; Soil contamination
• ISSN: 2064-5260; 2064-5279; 2064-5279
• DOI: 10.3311/PPee.37570

The amount of mushrooms cultivated around the world is constantly increasing, and the most commonly consumed species in Europe is the white button mushroom (Agaricus bisporus). Mushroom producers are facing a permanent challenge to provide the labour for harvesting, with increasing wage demands. Due to high market quality requirements, early automatized technologies are currently not able to replace manual picking. Our research therefore aims at facilitating the automated picking of button mushrooms and improving the technology via image processing. We aim to develop a method that can select the right size of mushrooms from field images and produce their picking position. We used Python programming language, along with the OpenCV and NumPy libraries, to implement image processing on real scenario images. The development considered factors such as fused- or overlapping mushroom heads, emergence of mushrooms from under caps, fallen or laterally visible stumps, cover soil contamination, and white mycelia which make detection significantly more difficult. We managed a solution for handling fruiting bodies that extend beyond the edge of the image due to the small field of view. The results indicated that the quality of photographs is crucial for the program's performance, as improper lighting, the presence of shadows. The efficiency of the algorithm was significantly affected by the 82% accuracy of the OpenCV Watershed segmentation algorithm, which in some cases could not separate objects. The program processed the images at an average speed of 0.78 seconds and produced the coordinates with a 92% success rate.