FTP Students Create Algorithm using YOLOv8 for Cocoa Defective Seeds Detection

The fact that Indonesia was ranked 3rd in the world in terms of cocoa exports which was the beginning of the research journey begins. One of the mandatory things done before the cocoa export is the sorting process. The sorting process carried out until now is still carried out conventionally, so it is very ineffective in terms of time and sorting results. Therefore, through the Agritech Science Innovation and Competition (ASIC) program organized by Agritech Research and Study Club Faculty of Brawijaya University, Yolkabrow Team made a new breakthrough, “Implementation of Yolov8 algorithm in detection of cocoa defective seeds to increase cocoa exports in Indonesia”

Their research was carried out for five months from June to November. This research was conducted by Yolkabrow Team consisting of 5 students of the Faculty of Agricultural Technology, including Nurhikmah Makmur, Aldina Hikmaktus Sakdiyah, Fransiskus Rio Pandi, Gyan Permata Aulia, Roy Ardy Colas Napitupulu. The research process was carried out with the guidance of lecturer Dr.Agr.Sc.Ir. Dimas Firmanda Al Riza, ST., M.Sc, IPM.

The use of Yolov8 which is one of the branches of the Convolutional Neural Network (CNN) is proven to be accurate in detecting cocoa defective seeds. The algorithm they make can be used in the detection of cocoa beans quickly and real time, they can also prove that the results of this algorithm have a high level of accuracy.

“The detection carried out by this algorithm uses differences in color and texture in the seeds of each type of cocoa bean defect. Where the type of cocoa defective seeds can be classified into three namely slaty seeds, moldy seeds, and germination seeds and we also use normal cocoa beans as a comparison. To see the quality of cocoa beans, it still needs destructive with a cutting test since the quality of the seeds can only be seen on the inside of the cocoa beans, “Nurhikmah said.

After they did a trial and error using several variations of the epoch used, they finally got the optimal algorithm results. The existence of this innovation is expected to help the cocoa industry sector in terms of detection of cocoa beans before export and production processing. The journey of this idea is based on two things, Indonesia, which is ranked third in cocoa exports, as well as the sorting process that is still carried out conventionally.

“Until now the sorting of cocoa beans is still carried out conventionally, namely by viewing one by one cocoa beans using the eyes, the sorting process can take a long time and cause bias from each person when sorting. Therefore, we consider conventional sorting very ineffective, “explained Nurhikmah

The Yolkabrow team succeeded in creating a cocoa bean detection algorithm with a high level of accuracy. The algorithm obtained using EPOCH is 150 and Batchsize 25. The results of this trial and error produced results until the mean level of average precision (MAP) was 99.3%. This study has also proven that the algorithm innovation created can recognize each cocoa defect seed with a percentage of 90%confidence.

“The way our own algorithm works when the camera captures the image of cocoa samples, in real time on the monitor, the sorted cocoa defective seeds will be immediately detected. Classification of Defective Seed Types will appear per each seed and there are numbers that indicate the level of confidence of this algorithm in detection. After trying directly on the cocoa defective seeds, 90%confidence level has been obtained, “Nurhikmah’s explanation. (*/Oky/UB PR/ Trans. Iir).