DETEKSI KEMATANGAN TANDAN BUAH SEGAR (TBS) KELAPA SAWIT BERDASARKAN KOMPOSISI WARNA MENGGUNAKAN DEEP LEARNING

Classification of oil palm fresh fruit bunch (FFB) based on maturity is very important for estimating oil content. Traditional methods using human vision to observe color changes during ripening and counting the number of fruits that fall from FFB are not effective. Research for neural architectures to design new network bases and improve them resulted in a set of models called EfficientNet. The most important function is the optimizer. This function repeatedly increases the parameters to reduce loss. In this study, the EfficientNetB0 and B1 models were developed to detect oil palm maturity into 6 classes, Raw, Ripe, Overripe, Underripe, abnormal, and empty bunch using optimizer RMSprop and SGD. From the research results, obtained the highest accuracy using the RMSprop optimizer of 0.9955 using the EfficientNetB0 model and 0.9949 using the EfficientNetB1 model. While using the SGD optimizer, the accuracy achieved is 0.918 using the EfficientNetB0 model and 0.9079 using the EfficientNetB1 mode