Detection of Tomato Plant Diseases Using Deep Learning

Chua, Yu Yang (2022) Detection of Tomato Plant Diseases Using Deep Learning. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Chua Yu Yang.pdf Restricted to Registered users only Download (5MB)

Abstract

Tomato plants diseases have been threatening the quantity and quality of tomato production for a few centuries. Several types of microorganisms, such as fungus, bacterial etc. have caused a variety of tomato leaves diseases that spread from the leaves to the tomato fruits. Therefore, precaution is necessary during the early state to detect and cure the diseases that occurred on the leaves before the diseases spread to result in plants death. However, the detection method employed currently is inconsistent and inefficient. Most farmers employ manual visualisations to identify the types of plant diseases as the automatic monitoring and processing of diseases cannot be achieved. Nonetheless, naked-eyes inspection is often inconsistent due to human factors such as fatigue, environmental factors and most importantly the varying professional experiences with leaves diseases. Such obstacles drive the need for a plant diseases detection system, aided with artificial intelligence (AI). Therefore, in this work, four different Convolutional Neural Network (CNN) models for tomato plant diseases detection, namely VGG-16, ResNet-50, DenseNet-121 and ResNet-101 were implemented and the hyperparameters were manipulated for parametric studies. The performance of the different architecture with the selected hyperparameters was evaluated and compared according to the well-established evaluation metrics for deep learning; Precision, Recall, F1 score, Matthews correlation coefficient (MCC) and Accuracy. The system was created by using Python programming language and executed in Google Colab using images obtained from the Plant Village dataset which consists of 10 classes: 9 types of tomato leaf diseases and a category of healthy tomato leaf. Based on accuracy and convergence, the selected hyperparameters are 0 of dropout layer, 50 of training epochs, SGD optimizer with 0.9 momentum rate and 0.001 of learning rate. The highest accuracy in tomato leaf diseases categorization was 100%, achieved by models ResNet-50, DenseNet-121 and ResNet-101 that employ augmented datasets. These CNN architectures have also presented excellent precision, recall, F1 score and MCC. Nonetheless, the ResNet-50 model required a shorter training time compared to DenseNet-121 and ResNet-101 models to achieve good performance. Performance of VGG-16 is less superior, reporting performance matrices that are about 90%, suggesting that a deeper network performs better than a shallow network in tomato leaf disease classification. Combining all its merits, the ResNet-50 model is proposed as the final model to be implemented in the tomato plant diseases classification system. In short, a CNN model that is capable of effective detection and classification of tomato leaf diseases has been developed and its benefits for tomato cultivation is much anticipated.