ORANGE PLANT LEAF DISEASE DETECTION AND CLASSIFICATION WITH IMAGE PROCESSING USING A DEEP CONVOLUTIONAL NEURAL NETWORK

Abstract

The farming of citrus is a crucial component of Pakistan’s fruit-based agricultural economy. But, the foliar diseases citrus canker, black spot, and greening have been posing a constant threat on citrus’s productivity. An optimal solution is an early and accurate detection of these diseases to improve the productivity. Therefore, this paper proposes an automated citrus leaf disease detection and classification framework based on deep convolutional neural networks (DCNNs). The proposed solution has five stages: image acquisition (dataset), preprocessing, data augmentation, deep feature extraction and optimization, and disease classification. Firstly, the images are obtained from a public dataset downloaded from Kaggle. Secondly, preprocessing techniques are used to improve the image quality and shape, thirdly the data augmentation techniques are used to enhance the model generalization, fourthly pre-trained models DenseNet-121, MobileNet, and InceptionV3 with transfer learning technique to extract deep features, and finally Adam optimizer and categorical cross-entropy loss function are used to fine tune the pre-trained models for classifications. The proposed model is evaluated on accuracy, precision, recall, and F1-score metrics. All the models demonstrated robust performance while DenseNet-121 achieved the best performance. The evaluation results assured the robustness of the use of transfer learning-based DCNN in citrus leaf disease detection.

Plant phenotyping relevance

柑橘葉の病害状態を画像から直接検出・分類する深層学習ワークフローが研究の中心であり、植物病害フェノタイピング手法に該当する。

Code and data availability

The paper's phenotyping input is a public Kaggle citrus leaf image dataset (654 RGB images of healthy, blackspot, canker, and greening leaves) explicitly cited with a URL matching an allowed URL. No author code or trained models are reported as publicly available.