Abstract:
Tomato is also a profitable crop in Bangladesh and many other countries because
of its high price and market value. But the development and productivity of tomato are
constantly threatened by several kinds of leaf diseases, which rapidly damage these plants.
The early diagnosis of plant illness and continuous monitoring of plant health are important
for the removal of factors that affect crop yield. This work presents an automatic way to
detect and recgnize tomato leaf diseases by a novel deep learning mechanism. For tomato
forest-based diseases, we create the custom dataset which has 600 labeled images includes
150 images for Tomato Early Blight, 150 images for Tomato Late Blight, 150 images for
Tomato Leaf Mold, and 150 images are healthy for leaves. These images were captured
directly from local farm fields to be as practical as possible. Four pretrained CNN models
VGG16, VGG19, ResNet50V2, and InceptionV3 were tested to find the optimum
performing model based on transfer learning. VGG19 and InceptionV3 obtained the
maximum accuracy of 97% from them. This technology can assist farmers to quickly and
correctly identify crop diseases, make timely corrective measures, therefore enhancing the
quality and yield of the crops, especially referring to remote rural areas where expert
consultation is not easily accessible.
