Chili Object Detection with Automatic Labeling and Mask R-CNN Model

To reduce labor-intensive work, we need to adopt autonomous robots in the agricultural sector.The process of traditional chili fruit harvesting requires a workforce and needs to improve for grading due to human eyes are sometime see the errors.In addition, the characteristics of chili fruits are not the same other type of fruits due to its sizes, variations, texture and its localization from the plant.To solve this issue, we conducted chili object detection using an automatic labeling algorithm and the Mask R-CNN model to localize and classify chili fruit variations.An automatic labeling algorithm is proposed for effectively exporting feature JSON files instead of pre-trained images using image annotation tools with manual processes.In our experimental results, automatic labeling detected chili objects with polygon points, and an accuracy rate of 85% was obtained with 4200 training images and 1800 testing images.The chili detection accuracy rate of 95% is tested with the Mask R-CNN model, and 2100 training images, 2100 validation images, and 1800 testing images are used in this paper.