Guava Fruit Detection and Classification Using Mask Region-Based Convolutional Neural Network

Abstract

Guava has various types and each type has different nutritional content, shapes, and colors. It is often difficult for some people to recognize guava correctly with so many varieties of guava on the market. In industry, the classification and segmentation of guava fruit is the first important step in measuring the guava fruit quality. The quality inspection of guava fruit is usually still done manually by observing the size, shape, and color which is prone to mistakes due to human error. Therefore, a method was proposed to detect and classify guava fruit automatically using computer vision technology. This research implements a Mask Region-Based Convolutional Neural Network (Mask R-CNN) which is an extension of Faster R-CNN by adding a branch that is used to predict the segmentation mask in each region of interest in parallel with classification and bounding box regression. The system classifies guava fruit into each category, determines the position of each fruit, and marks the region of each fruit. These outputs can be used for further analysis such as quality inspection. The performance evaluation of guava detection and classification using the Mask R-CNN method achieves an mAR score of 88%, an mAP score of 90%, and an F1-Score of 89%. It can be concluded that the proposed method performs well in detecting and classifying guava fruit.