Skripsi
MODIFIKASI YOU ONLY LOOK ONCE-CLS DENGAN BACKBONE EFFICIENTNET UNTUK KLASIFIKASI GANGGUAN PADA DAUN KELAPA SAWIT
Oil palm is a strategic global commodity that plays an important role in the global vegetable oil supply. Its productivity is often disrupted by leaf attacks such as dryness, fungal disease, magnesium deficiency, and scale insects. Early detection is necessary to prevent the spread of pests and minimize losses. Manual diagnosis by agronomists is time-consuming and inefficient on a large plantation scale, making the use of deep learning a solution for automatic leaf disorder classification. YOLO-cls is a development of YOLO that focuses on image classification. The performance of YOLO-cls is influenced by the selection of the backbone architecture. EfficientNet was developed using a compound scaling approach and MBConv blocks to balance depth, width, and resolution, ensuring efficiency and high performance. This study compares the variations of EfficientNet-B (B0–B7) and EfficientNet-V2 (V2S, V2M, V2L) as backbones in YOLO-cls to obtain the best variation in palm leaf disease classification. Performance evaluation is carried out using precision, sensitivity, F1-score, specificity, and accuracy. The results showed that EfficientNet-V2S as the YOLO-cls backbone achieved the best performance with a precision of 97.04%, sensitivity of 97%, F1-score of 96.88%, specificity of 98.99%, and accuracy of 98.5%. These results indicate that selecting the appropriate EfficientNet variant significantly improves classification performance for oil palm leaf disorders.
No other version available