Comparison of MobilenetV2 and EfficiennetB3 Method to Classify Diseases on Corn Leaves

Abstract

Corn is an important commodity in Indonesia and one of the world’s staple foods. According to FAO in 2017, disease problems often hamper corn production. The main problems that affect crop yields are diseases that damage corn leaves, including rust, spot, blight and downy mildew. Manual disease identification on corn leaves has limitations in consistency and scalability. A digital image processing system is needed to increase the speed and accuracy of recognition. Classification of types of corn leaf disease is needed so that farmers can distinguish types of corn leaf disease. Therefore, to identify four forms of maize leaf disease, this study tested a CNN model using the EfficientNetB3 and MobileNetV2 architectures. The data used to evaluate these two models is divided into training and testing subsets. Based on test results with 50 epochs, EfficientNetB3 obtained 93.20% accuracy with a loss of 0.0850, while MobileNetV2 obtained 92.48% accuracy with a loss of 0.19020. When the test data is evaluated, EfficientNetB3 performs better than MobileNetV2. MobileNetV2 has limitations in handling complex feature representation on very complex data. On the other hand, although it provides better results, EfficientNetB3 has the disadvantage of high computing and memory resource requirements, which can hamper efficiency in practical implementation. Based on these findings, EfficientNetB3 is recommended because it performs better in maize leaf disease classification, shows smaller losses and higher accuracy than MobileNetV2. By using this model, corn plant diseases can be identified efficiently and precisely, thereby improving disease control and producing more productive corn plants.