Apple Leaf Disease Classification Using Deep Learning Interim Assignment

1. Introduction and Overview

Background and Research Gap

The agricultural sector is determinant in food security worldwide and diseases of plants remain one of the key threats to production and product quality. Apple trees are economically and nutritiously important trees that are prone to many leaf diseases comprising of Alternaria leaf spot, Rust, Brown spot, and Gray spot. The detection of the diseases traditionally involves manual examination of agricultural specialists, a lengthy and inaccurate procedure. The use of automated approaches based on deep learning has been an option since these methods are accurate and scalable. Although the use of deep learning in plant pathology has been proved in earlier research, there is a need in comparing the effectiveness of using custom Convolutional Neural Networks (CNNs) with transfer learning in apple leaf classification (Alsayed et al., 2021). This study fills this gap through the implementation and evaluation of both strategies to a preselected apples leaf disease data set.

Native Assignment Help provides academic guidance for students working on artificial intelligence, machine learning and deep learning projects. An experienced Assignment Helper UK can help you understand CNN architectures, transfer learning, image preprocessing and model evaluation techniques. Our support helps students structure technical reports, explain research findings and present complex computing concepts clearly in line with UK academic expectations.

Specification

The proposed project will build an image-based classification system based on deep learning to classify the apple leaf diseases. Research questions will be measuring the accuracy of the classification of custom CNN training over the transfer learning types, finding good preprocessing and augmentation work and testing the actual deployment possibility in the real world (Khan et al., 2021). The used tools and techniques are Python, PyTorch, torchvision, NumPy, matplotlib, and Seaborn.

Ethical, Legal, Social, and Professional Issues

This project also follows ethical procedures since it is based on open licenses or settlement only using publicly available datasets. The distribution of the data was downloaded at Kaggle and there is no inclusion of personal sensitive information. The ethical procedure in the university indicates that no human or animal subjects requiring an ethical approval were involved and there is no conflict over the ownership of the data. At the professional level, the system is displayed as supportive, not alternative, to agronomists, which enhances reasonable use of AI. Credit to past academic work has been given by means of in-text citations in order to maintain academic integrity.

2. Literature Review

Deep learning has depicted immense potential in agriculture applications in the detection of plant diseases face. Jiang et al. (2019) built a better CNN model to detect apple leaf diseases and presented the high levels of accuracy, which also characterises the effectiveness of the model under the real-time conditions. Khan et al. (2022) designed a diagnostic system of apple leaves diseases based on the methods of deep learning, implementing it into a real-time architecture. Chao et al. (2020) compared a number of architectures and decided that transfer learning dramatically enhanced classification in limited databases. In a similar way, the transfer learning based on ResNet was applied by Ozden (2021) who showed better results in terms of accuracy and shorter training time better than CNNs trained in a supervised manner.

These are some proofs about the significance of the choice of the architecture and the efficiency of the transfer learning that is exactly the focus of this project. The literature also focuses on the importance of preprocessing, and augmentation techniques to increase the model generalization, and the current project has touched upon normalization, resizing, and transformation of data.

3. Progress to Date

The data acquisition, preprocessing and exploratory data analysis have made significant progress. The dataset is retrieved form Kaggle and consists of 1,641 images that are divided into the five classes of Alternaria leaf spot (278 images), Brown spot (215), Gray spot (395), Healthy leaf (409), and Rust (344). Images were loaded and transformed with torchvision.datasets.ImageFolder method through standard preprocessing operations as resize to 224x224 pixels, convert to tensors and standardization.

The analysis of a class distribution was carried out with Seaborn, which pointed to the lack of equivalence in classes. This imbalance understanding will be used in further augmenting the data to avoid model bias. To ensure the image quality and assignment to the correct classes, there is a grid visualization of the sample images of each class. The visualizations prove that the dataset can be used in classification activities by using deep learning.

Environmental configuration of Python, loading of data, and confirmation of the classes have been done, as well as plotting numbers of the classes and samples. All these are preparatory steps to the training and testing of custom CNN and transfer learning model ResNet50. The methodology adheres to the common supervised learning pipelines in which the data will be partitioned into the training and validation set, as well as the test set. To make sure that the models are assessed thoroughly, the metrics of accuracy, precision, recall and F1-score will be used.

4. Planned Work and Project Plan

The project started with the identification and the acquisition of datasets, continued with the exploratory analysis of the data, and preprocessing. Such initial steps were loading the dataset, ensuring class label, image resizing, and normalization, and distribution, and sample visualization of classes. The second stage is the model development during which a custom CNN will be built and trained completely to categorise apple leaf diseases. Simultaneously, transfer learning will be executed with pre-trained ResNet50 model and made suitable and relevant to this task by fine-tuning. The processed set of data will be used to train the models and will be tested based on accuracy, precision, recall, F1-score and confusion matrices. Generalization will be enhanced through data augmentation by randomly rotating, flipping and brightness manipulating the data. The final phases include conducting a serious comparison of the two performances, reporting of the results, and completing the final report and presentation as well as the demonstration. 

The project quality will be measured depending on the evaluation metrics, proper documentation, and lessons learned by comparing the models. Final assessments there will not only be presented quantitative ones, but there will also be presented qualitative including the generalization of the real world feasibility.

Reference List

Alsayed, A., Alsabei, A. and Arif, M., 2021. Classification of apple tree leaves diseases using deep learning methods. International Journal of Computer Science & Network Security, 21(7), pp.324-330.

Chao, X., Sun, G., Zhao, H., Li, M. and He, D., 2020. Identification of apple tree leaf diseases based on deep learning models. Symmetry, 12(7), p.1065.

Jiang, P., Chen, Y., Liu, B., He, D. and Liang, C., 2019. Real-time detection of apple leaf diseases using deep learning approach based on improved convolutional neural networks. Ieee Access, 7, pp.59069-59080.

Khan, A.I., Quadri, S.M.K. and Banday, S., 2021. Deep learning for apple diseases: classification and identification. International Journal of Computational Intelligence Studies, 10(1), pp.1-12.

Khan, A.I., Quadri, S.M.K., Banday, S. and Shah, J.L., 2022. Deep diagnosis: A real-time apple leaf disease detection system based on deep learning. computers and Electronics in Agriculture, 198, p.107093.

Özden, C., 2021. Apple leaf disease detection and classification based on transfer learning. Turkish Journal of Agriculture and Forestry, 45(6), pp.775-783.