Apple Disease Dataset

Description

This dataset contains images of Manalagi apple varieties from Indonesia. The data represent both healthy and diseased fruit. All data were collected from apple orchards owned by farmers. The dataset consists of four categories: one healthy fruit class and three disease classes: Anthracnose, Black Pox, and Powdery Mildew. Images were captured using a digital camera and a smartphone. To increase the quantity and variety of data, image augmentation was performed. The augmentation techniques used included rotation (45°, 90°, 180°, 225°, and 270°), flipping (vertical and horizontal), noise addition (Gaussian noise, speckle noise, and salt-and-pepper noise), and image transformations such as vertical and horizontal shifts, saturation changes, and brightness adjustments. This process aims to enrich the data variety so that the model can learn more robustly to various field conditions. Dataset Summary: - Total original data: 579 images - Total data after augmentation: 8,396 images - Image resolution: 1024 × 1024 pixels Data Distribution: - Healthy: 102 → 1,530 images - Anthracnose: 163 → 2,445 images - Black Pox: 166 → 2,201 images - Powdery Mildew: 148 → 2,220 images Dataset Description: - This dataset focuses on fruit diseases (fruit-based dataset), which is still rare compared to public datasets that generally focus on leaves. - The data is collected directly from orchards under natural conditions, reflecting real world variations such as lighting, background, and disease manifestations. - Augmentation techniques help improve the performance of deep learning models by expanding the data variety and reducing the risk of overfitting. - This dataset can be used as a benchmark for training, validating, and evaluating apple disease classification models. - This dataset also supports the development of automated disease detection systems, smart farming, and AI based applications in agriculture.

Steps to reproduce

Data Collection Images of apple fruit were collected from multiple apple orchards located in major apple production areas. The dataset includes images of both healthy fruit and fruit affected by common apple diseases. Data collection was conducted directly in orchard environments to ensure that the images reflect real field conditions. Field Verification Consultations were carried out with local farmers and agricultural experts to identify the most frequently observed disease symptoms on apple fruit. This process ensured that the selected disease categories accurately represent conditions commonly encountered in the field. Image Acquisition Image acquisition was performed using a combination of digital cameras and smartphone devices with different camera specifications, including a DSLR camera and smartphones such as Realme 9 Pro (16 MP main camera), Apple iPhone 13 (12 MP dual camera), and Samsung Galaxy S Series (64 MP main camera). All images were captured under natural lighting conditions in orchard environments, without the use of artificial illumination. Data Validation and Categorization The collected images were reviewed with the assistance of agricultural experts to verify their accuracy. Each image was then classified into one of five categories, consisting of one healthy fruit class and four apple fruit disease classes, based on clearly observable visual symptoms. Data Preparation The final dataset was organized into separate folders corresponding to each category. All images were stored in JPG format with a consistent resolution and standardized structure, making the dataset readily usable for machine learning and computer vision applications.

Institutions

• Bina Nusantara University

  DKI Jakarta, West Jakarta

Categories

Licence