Annotated Banana Ripeness Stage Dataset

Description

This dataset consists of high-resolution images of bananas categorized into three distinct ripeness stages: unripe, ripe, and overripe. Each image contains manual bounding box annotations to support object detection and classification tasks in agricultural computer vision. The project was conducted under the supervision of Md. Abu Raihan, with contributions from the research team regarding data collection and image labeling. This dataset is intended for training machine learning models to automate quality assessment in the food supply chain. Dataset Specifications: Primary Subject: Musa acuminata (Banana) Total Image Count: 819 Classes: Unripe, Ripe, Overripe Annotation Type: Bounding Boxes Dataset Format: YOLO v8 Data Split Ratio: 75% Train | 15% Validation | 15% Test Training Set: 573 images Testing Set: 125 images Validation Set: 121 images

Steps to reproduce

Phase 1: Data Acquisition Image Collection: Capture or source high-resolution images of bananas (Musa acuminata). Environmental Diversity: Ensure images include various lighting conditions and backgrounds to prevent model overfitting. Subject Selection: Group bananas into three physical categories: Unripe: Solid green peel. Ripe: Yellow peel with minimal green. Overripe: Yellow peel with brown/black senescent spotting. Phase 2: Annotation and Labeling Tool Selection: Use an annotation tool (e.g., CVAT, LabelImg, or Roboflow). Bounding Box Creation: Draw tight rectangular boxes around each individual banana in the image. Class Assignment: Assign each box a specific class index: 0: Unripe 1: Ripe 2: Overripe Export Format: Export all annotations in YOLO v8 format (normalized .txt files containing class_id x_center y_center width height). Phase 3: Dataset Partitioning (Splitting) Shuffle: Randomize the images to ensure no bias in the training set. Distribution: Separate the files into three distinct folders using the specific counts: Train: 80% images + labels. Test: 10% images + labels. Validation: 10% images + labels. Structure Verification: Ensure each image filename (e.g., banana_01.jpg) matches its corresponding label file (e.g., banana_01.txt). Phase 4: Configuration Setup Create YAML: Define the data.yaml file to point to the train, val, and test directories. Define Classes: Set the nc (number of classes) to 3 and list the names: ['unripe', 'ripe', 'overripe'].

Institutions

• Khwaja Yunus Ali University

  Rajshahi Division, Chauhāli

Categories

Licence