Banana_Tree_Disease_Detection_Dataset(BTDDD)
Description
The dataset is a collection of images representing various conditions of bananas, specifically aimed at training machine learning models for image classification or augmentation tasks. The dataset is organized into multiple subfolders, each representing a different condition or class of bananas. These classes include: Healthy Bananas Bananas with Fusarium Wilt Bananas with Natural Leaf Death Bananas with Rhizome Root Issues Each image in the dataset is initially stored in its respective class folder and typically contains a banana or bananas under different conditions, viewed from different angles, and possibly with varying levels of resolution or lighting. The dataset is then processed for various machine learning tasks like classification, detection, or augmentation. Specifically, this dataset is aimed at providing a variety of augmented images to ensure a more robust training set, which is critical for improving the generalization performance of machine learning models. Note for Researchers Using the dataset This dataset was created by Shuvo Kumar Basak. If you use this dataset for your research or academic purposes, please ensure to cite this dataset appropriately. If you have published your research using this dataset, please share a link to your paper. Good Luck.
Files
Steps to reproduce
Resizing: All images are resized to a consistent resolution of 256x256 pixels to ensure uniformity and ease of processing in machine learning models. This resizing step helps in standardizing image dimensions for input into neural networks. Augmentation: Image Augmentation techniques are applied to increase the variety of the dataset without needing to gather additional images. These augmentations are crucial to prevent overfitting and ensure that the model learns from a wide variety of transformations. The following methods are applied to each image: Horizontal Flip: The image is flipped horizontally to simulate different orientations of bananas. Vertical Flip: The image is flipped vertically to add variation in the orientation of bananas. Random Rotation: Each image is rotated randomly between 10 to 30 degrees to simulate different angles. Brightness Adjustment: The brightness of the image is modified randomly between 0.5 to 1.5 times, making the dataset more robust to lighting conditions. Contrast Adjustment: Random contrast adjustments are made between 0.5 and 1.5 times to simulate various lighting and exposure conditions. Color Adjustment: The image's color saturation is adjusted randomly to simulate different environmental factors. Gaussian Blur: A Gaussian blur is applied to simulate out-of-focus or low-quality images. Resize: Random scaling of the image size is applied between 80% and 120% of the original size to simulate different object sizes. Shear Transformation: The image is sheared randomly between -0.2 and 0.2 to simulate image distortions. Invert Colors: The colors in the image are inverted to create different visual effects that can be useful for training models that are resilient to color variations. File Naming and Saving: The augmented images are saved with new filenames in the respective class subfolders. The filenames are numbered incrementally (e.g., 1.jpg, 2.jpg, etc.), ensuring that the images are uniquely identified. The augmented images are stored in a new directory structure under a separate folder named Augmented_Banana_resized2. This structure mirrors the original dataset's subfolder structure to preserve the class labels and facilitate easy tracking of augmented data. Handling Unsupported or Invalid Files: Files that are not in supported image formats (such as .jpg, .png, .jpeg, .bmp) are skipped during processing. The dataset is also checked to ensure that files are valid images using the imghdr library, which confirms the actual format of the images. Outcome of the Procedure: The dataset is significantly expanded by the augmentation process, providing a diverse set of images that help improve the performance and robustness of machine learning models. The augmented images are now suitable for training models that can handle variations in lighting, orientation, scale, and quality of banana images, which is essential for real-world applications like disease detection, classification, or fruit quality inspection.