Orange Fruit Diseases Dataset
Description
The dataset contains dominant diseases affecting oranges such as Blacks spot, Canker, Greening, Scab and a Healthy one. The dataset can be used for any artificial intelligence project and any agricultural research for precision Agriculture. This dataset was carefully curated from https://data.mendeley.com/datasets/3f83gxmv57/1 and https://www.kaggle.com/datasets/jonathansilva2020/orange-diseases-dataset repository respectively for the purpose of research works in computer vision. To enhance robustness and generalizability, we gathered these datasets from both repos. This approach allowed us to include images with varying lighting conditions and qualities, thus strengthening the model’s validity across diverse real-world scenarios.
Files
Steps to reproduce
Dataset Curation The manuscript details the creation of a dataset specifically tailored for the study: - Scope: A curated dataset of 700 annotated images of diseased and healthy oranges was used, covering conditions like black spots, cankers, greening and scab — prevalent diseases affecting orange fruits. The dataset was obtained from the Mendeley and Kaggle dataset repository respectively. - Annotation Process: Annotation was performed using bounding boxes to mark regions of interest, ensuring high-quality labels for training and evaluation. Annotation was done using Cvat annotation tool. The annotated files were then converted into a Yolo format. Setting up the python environment: sudo apt update sudo apt install python3-venv # Installing python3-venv if not installed python3 -m venv yolov11_custom-env source yolov11_custom-env/bin/activate pip install --upgrade pip pip install torch torchvision opencv-python Cloning the YOLOv11 Repository from github git clone https://github.com/ultralytics/yolov11.git cd yolov11 pip install -r requirements.txt Creating directory and uploading the dataset via SFTP Using the Termius's SFTP client to upload the dataset to the server. ../home/ben/Yolov11_Custom mkdir -p datasets/ home/ben/Yolov11_Custom /images mkdir -p datasets/ home/ben/Yolov11_Custom /val Training the Model Parameter adjustments Downloading Yolov11nano from github Initial training result Introducing an Optimizer ( Adamax Optimizer) Adamax is an optimizer that builds upon the Adam optimization algorithm. It uses the infinity norm (maximum) of the gradient instead of the L2 norm. This provides superior stability and robustness against sparse gradients, making it suitable for tasks with noisy or irregular updates. The Adamax optimizer in YOLOv11 was modified in the training script and configuration files.