Aminuddin, Nuramin Fitri (2023) Development of chilli leaf disease identification using convolutional neural network. Masters thesis, Universiti Tun Hussein Onn Malaysia.

Preview	Text 24p NURAMIN FITRI AMINUDDIN.pdf Download (690kB) | Preview
	Text (Copyright Declaration) NURAMIN FITRI AMINUDDIN COPYRIGHT DECLARATION.pdf Restricted to Repository staff only Download (325kB) | Request a copy
	Text (Full Text) NURAMIN FITRI AMINUDDIN WATERMARK.pdf Restricted to Registered users only Download (26MB) | Request a copy

Abstract

Over the years, numerous studies have been conducted on the integration of computer vision-based feature descriptors and machine learning classifiers for crop disease identification to help farm owners who need assistance in monitoring crop health. However, these conventional feature descriptors often require manual extraction of various disease features. Improperly extracting disease features that are discriminative from the crop can jeopardise the identification performance of the classifiers. To overcome the limitation of these descriptors, a deep convolutional feature descriptor, namely Convolutional Neural Network (CNN), is implemented in this research for chilli leaf image-based disease identification. Three CNN-based models, namely DenseNet-201, EfficientNet-b0, and NasNet-Mobile, are used in this research. Healthy and diseased chilli leaf images are collected and have their resolution resized prior to the model training. The modified model version of DenseNet-201, EfficientNet-b0, and NasNet-Mobile are also built, with the classification layer of each model (softmax-based) is replaced by a Support Vector Machine (SVM) based layer. The identification performance of DenseNet-201 (softmax-based), EfficientNet-b0 (softmax-based), and NasNet-Mobile (softmaxbased) are compared with their modified variants, namely DenseNet-201 (SVMbased), EfficientNet-b0 (SVM-based), and NasNet-Mobile (SVM-based). It is found that the EfficientNet-b0 (SVM-based) model has outperformed the rest of the models with the highest identification performance, where the performance index of accuracy, recall, specificity, precision, and F1-score being 97.33%, 0.97, 0.94, 0.95, and 0.96, respectively. Additionally, a chilli leaf image-based disease identification system in the form of MatlabTM Graphical User Interface (GUI) with captured image input and identification result terminal has been developed in the research to assist farm owners and non-chilli experts in identifying chilli diseases. The EfficientNet-b0 (SVM-based) model is deployed as the core of the developed GUI

Actions (login required)

View Item