Loading...
Loading...
Contributors:
Dataset URI:
10.5281/zenodo.23494150This dataset present a new fish imagery called Patin Pekan River-Fish (PPR-Fish) for machine learning applications. The dataset was collected from fish videography recorded in Malaysia’s Pekan River, located near the South China Sea. Dataset Collection, Processing, and Annotation The PPR-Fish dataset was collected from river-based cage aquaculture farms along the Pekan River in Pahang, Malaysia, focusing on Patin fish (Pangasius spp.). Data collection was conducted between August and October 2024 using an iPhone 15 camera with a 48-megapixel imaging capability. The videos were recorded under real-world riverine aquaculture conditions, where the water was naturally brown and turbid. A total of 12 videos were recorded to capture fish appearance and activity under these environmental conditions. The recorded videos were subsequently converted into individual frames at a sampling rate of two frames per second to construct the image datasets for fish detection and activity classification. For the fish activity classification dataset, the extracted frames underwent five types of data augmentation, resulting in a total of 9,099 frames. Following quality inspection, 6,500 frames were selected for the final classification dataset, while frames exhibiting substantial blur or other image-quality issues were excluded. The selected frames were assigned to two classes: active and not active. The active class represents fish exhibiting visible activity, such as swimming rapidly or aggressively near the water surface and feeding zone, whereas the not active class represents fish with limited visible movement or fish that were submerged and not visibly active near the surface. These labels were based on observable visual characteristics rather than direct physiological measurements of hunger. The classification dataset was divided into training, validation, and testing subsets using a 70:20:10 ratio, respectively, to support model training, hyperparameter validation, and independent performance evaluation. For the fish detection dataset, 3,250 frames were selected from the extracted frames for object detection and annotation. The LabelImg annotation tool was used to manually annotate visible fish using bounding boxes, with a single class label, fish. The annotations identify the locations of fish within the images and provide the corresponding labels required for training object detection models. The detection dataset was also divided into training, validation, and testing subsets using a 70:20:10 ratio, respectively. The classification and detection datasets serve complementary purposes: the classification dataset supports the categorisation of fish activity into active and not active classes, whereas the detection dataset supports the localisation of fish within images. Together, these datasets provide resources for developing and evaluating computer vision approaches for fish monitoring in real-world riverine aquaculture environments.
Keyword(s):
MeSH Term(s):
Grants and Funding:
No Grants or Funding Data Available
Associated Publication(s):
No Associated Publication Data Available
Supplementary Data:
No Supplementary Data Available
Related Dataset(s):
No Related Dataset Data Available
The National Library of Medicine (NLM) provides information about biomedical datasets. Inclusion on an NLM Web page does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: NLM Web Policies (nih.gov)