Abstract

Agriculture is an essential field that includes crop production, plant and fruit growing, and livestock. Plant disease is a significant challenge in agriculture, which can drastically impact crop production and lead to reduced productivity and potentially severe shortages. Hence, it is essential to detect plant diseases as fast as possible in order to start separating diseased ones from healthy ones. However, this process is arduous and challenging to accomplish manually. This paper shows a possible automation technique using EfficientNet CNN models. Gray-level, binarized, and color-level datasets were separately given in this study. The images in the dataset were resized. In order to increase the training data's variety, the input images underwent a horizontal flip. The data was rotated, helping the model in handling minor orientation variances. Randomized zoom was implemented to improve the model’s ability to recognize leaf images from varying distances and sizes. The highest training accuracy achieved is 99.81% with the EfficientNetB5 model at 50 epochs and a batch size of 128. The EfficientNetB0 model achieves the lowest training accuracy with 97.17% accuracy at 20 epochs and a batch size of 16, however. The highest testing accuracy achieved is 97.83% by the EfficientNetB7 model with 50 epochs and a batch size of 64; on the other hand, the lowest testing accuracy is 77.84% by the EfficientNetB4 model with 20 epochs and a batch size of 32.

Keywords: Leaf disease, EfficientNet, CNNs, Computer vision, Tomato leaf disease, Plant diseases