Performance Comparison of CNN Models for Tomato Disease Detection using Image-based data in Both Controlled and Real-world Conditions

Meenakshi Thalor^1* , Yash Chavhan¹ and Sanjay Mate²

¹Department of Information Technology, AISSMS Institute of Information Technology, Pune, India

²Department of Information Technology, Government Polytechnic, Daman, India

Corresponding Author E-mail:meenakshi.thalor@aissmsioit.org

Abstract:

Tomato plants are integral to worldwide agricultural production, yet they remain vulnerable to numerous diseases stemming from fungi, bacteria, and viruses. Prompt and precise identification of these ailments is vital for maintaining crop productivity and safeguarding food supplies. This paper consolidates insights from revolutionary machine learning (ML) and deep learning (DL) methodologies, particularly convolutional neural networks (CNNs), for identifying tomato plant diseases. Employing datasets like Plant Village and authentic field specimens, we evaluate model performance across diverse scenarios. Findings indicate that CNNs attain over 99% accuracy in controlled environments but face considerable obstacles in practical field applications because in many real-world applications the data can vary greatly due to environmental factors such as lighting conditions, weather, and seasonal changes. This paper explores three CNN architectures DenseNet, ResNet50 and VGG16 while offering approaches to improve model adaptability and expandability for RealWorld implementation.

Keywords:

Disease; Deep Learning; DenseNet Leaf; Tomato; VGG

[ HTML Full Text]