Annotated underwater fish detection dataset from pond environments
Description
Fish detection using computer vision is increasingly recognized as significant in aquaculture, particularly for monitoring and managing fish populations in small-scale settings. Traditional fish population assessment methods are invasive, labor-intensive, and often inaccurate, prompting the development of automated solutions using computer vision and deep learning. This dataset comprises annotated images of the Orange Chromide (Etroplus maculatus) fish species from ponds in Kolathur, Chennai. The videos were captured using a Crosstour CT9000 underwater camera at 60fps with 1-minute intervals. Recordings were made inside the pond at depths of less than 4 meters, covering a wide angle of approximately 135 degrees. The videos include various complexities such as occlusion, turbid water conditions, high fish density per frame, and natural lighting conditions. Keyframes were carefully extracted from these videos and formatted at 640x640 pixels in .jpg format. Each image was annotated in .txt format using the bounding box style. The annotated dataset comprises 586 images, divided into train (409), validation (118), and test (59) sets. This dataset offers a comprehensive and challenging set of images that reflects real-world conditions in South Indian ponds. The dataset can facilitate advancements in computer vision applications within aquaculture, including automated fish population monitoring, growth tracking, health assessment, and behavior analysis.