Detection and Measurement System for Button Mushrooms Using Convolutional Neural Network

Abstract

In Malaysia, the button mushroom is recognized as a vegetable with high nutritional value and is easy to cultivate. Monitoring mushroom growth requires farmers to regularly inspect their crops, which is time-consuming and inefficient. Hence, an automated detection and measurement system for button mushrooms has been developed using image processing techniques based on convolutional neural network (CNN) algorithm model known as YOLOv4. The algorithm was utilized to train the system using button mushroom images to create training models. The performance of the YOLOv4 models was evaluated across various iterations ranging from 1000 to 6000 iterations. The model with 2000 iterations demonstrated the most effective performance based on Recall, Precision, F1-score, Time and Mean Average Precision metrics. The model was used in a small-scale experimental setup to evaluate the button mushroom detection and measurement system’s performance. Based on the results obtained from the experiments, the detection and measurement system demonstrated high accuracy in locating the position of each button mushroom with only a 5% deviation error in predicting the size of each button mushroom.