Detection of Begomovirus Disease for Identification of Disease Severity Level in Tomato Leaves Using Convolutional Neural Network (CNN)

Tomato is one of the horticultural crops that have high demand in the community, which demands continuous increase in production and quality. On the other hand, cultivating and caring for tomato are not simple. The challenge frequently encountered by tomato farmers is the prevalence of diseases on tomato leave caused by bacteria, viruses, and fungus. If the diseases are not treated appropriately, these diseases will cause a decrease in production due to the poor quality of tomatoes. In this study, the optimized Convolutional Neural Network (CNN) model for classification and detection Begomovirus disease in tomato leave is proposed. The optimized model is obtained from the comparison of three types architectures: SSD Mobilenet V1 FPN, EfficientDet D1 and SSD Resnet 50 V1 FPN. For the purpose of this research, a dataset consisted of tomato leaves contracted with various stages of begomovirus has been collected and prepared to be used as training and validation. The dataset is grouped into 4 classes based on the severity level of the disease according to the scale of symptoms that occur in tomato leave. The Experimental results shows that the mAP value of SSD Mobilenet V1 FPN, SSD Resnet 50 V1 FPN and EfficientDet D1 model are 56.17%, 61.58% and 66.03%, respectively. The EfficientDet D1 model resulted in the highest mean Average Precision (mAP) value due to this model produced the lowest loss value of 0.1 compared to the other 2 models. The proposed model uses BiFPN method for feature extraction which resulted in higher accuracy compared to previous models which only used FPN. This performance was able to be achieved with much less number of layers.