Automatic Mango Leaf and Trunk Detection as Supporting Tool of Mango Pest Identifier (MPI)

As a popular fruit in Indonesia, mango is widely consumed in the nation. Because mango can be planted quite easy then almost every family owns a mango plant. It can be planted in the yard or on a large plantation. Mango is also a potential export product. Unfortunately, the presence of pests has made Indonesian mangoes rejected in several export processes. The number of mango pest types is almost 90 and the pests attack various parts of the mango plant. Due to the similarity between pests, the machine learning model cannot recognize them effectively. To localize the pest type based on the attacked area will make a specialised model for each part of the mango plant. The approach will increase the chance of the specific model to recognize the correct mango pest. Consequently, recognizing part of the plant is an important step of improved mango pest detection systems. Therefore, this study objective is to provide an identification tool for trunks and leaves as they are the most attacked areas. By adding this process, the image can be filtered properly before it is injected into the pest recognition module. The Faster R-CNN method is used to identify trunk and leaf images. The results of the study are 74% accuracy, 65,789% precision, and 100% recall.