Multi Layered Convolutional Neural Network in Classification of Different Tomato Diseases

Abstract

Tomato leaf diseases are a major danger to worldwide crop yields, necessitating early and accurate detection for successful management. In this study, we assess and compare two transfer-learned convolutional neural network backbones, Inception V3 and ShuffleNet V2 x1.0, for categorizing seven typical tomato leaf states. We compiled an open source dataset of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{1 8}, \text{0 8 8}$</tex> high-resolution pictures from the PlantVillage repository that included bacterial spot, early blight, late blight, leaf mold, Septoria leaf spot, powdery mildew, and healthy leaves. To imitate real-world unpredictability, all photos were reduced to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$224 \times 224$</tex> pixels and normalized. Random flips, rotations, zooms, and brightness changes were also applied. Researchers replaced each network's original classifier with a seven-way softmax head and fine-tuned only the final two convolutional blocks, as well as the new classifier, using Adam optimization categorical cross-entropy loss and early stopping. On a held-out test set, ShuffleNet V2 achieved 95 percent overall accuracy, while Inception V3 achieved 88 percent. Precision recall, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$F 1$</tex> analysis, and confusion matrices demonstrate that ShuffleNet V2 not only improves accuracy but also minimizes misclassifications between visually comparable diseases. Its lightweight architecture, tiny memory footprint, and quick inference make it perfect for use with smartphones and embedded devices. Our findings show that transfer learning with ShuffleNet V2 provides a strong, efficient, and practical method for on-site tomato disease diagnosis.

Topics & Concepts