Black Tea Powder Dataset for Fine-Grained Visual Classification

The Black Tea Powder Image Dataset is a balanced multi-device image dataset designed for fine-grained classification of black tea powder grades using computer vision techniques. The dataset consists of 3,346 RGB images collected using two different imaging devices; a digital camera (Sony A7III) and a smartphone (Samsung Galaxy A22), to capture sensor-level variability and enhance robustness for machine learning applications.

The black tea powder samples were sourced from PT Perkebunan Nusantara VII Rancabali Tea Plantation, West Java, Indonesia. However, the image acquisition process was conducted independently and was not governed by industrial grading standards or visual quality control procedures of the plantation. Variations in color and appearance across images arise primarily from imaging devices rather than intrinsic differences in tea quality.

The dataset is divided into two subsets based on the acquisition device:
1. Camera-based dataset: 1,672 images
2. Smartphone-based dataset: 1,674 images

Both subsets were collected under controlled conditions using a consistent acquisition protocol, including fixed background and similar shooting distance, to ensure comparability across devices. While the same lighting setup was used throughout the acquisition process, variations in image appearance were observed due to differences in imaging devices and operators.

The dataset covers eight classes of black tea powder:
Broken Mix (BM) – 418 images
Broken Orange Pekoe (BOP) – 432 images
Broken Pekoe (BP) – 410 images
Broken Tea (BT) – 416 images
Dust – 403 images
Fanning 2 – 405 images
Pekoe Fanning (PF) – 457 images
Powdery Dust (PW Dust) – 405 images

Images are organized in a hierarchical directory structure based on the acquisition device and class label. At the top level, the dataset is divided into two main folders corresponding to the image capture devices (Camera and Phone). Within each device-specific folder, images are further grouped into eight subfolders representing the black tea powder classes: BM, BOP, BP, Broken Tea, Dust, Fanning 2, PF, and PW Dust.