Video footage of fish and water quality variables in a fish farming scenario

This dataset contains measurements of water quality parameters along with behavior-related variables obtained from fish recognition and localization using machine learning techniques on video footage. The water quality data were collected using a Hanna HI98194 multiparameter probe, while the video footage was taken with a webcam connected to a Raspberry Pi 4. A total of 247 uniformly distributed frames from the video footage were manually annotated to train a YOLOv4 model for fish recognition and tracking. Using the trained model, the complete video footage was processed and the time series of fish pairwise distance and average traveled distance were calculated from the recognition model. This dataset can potentially be used to study fish behavior-related and water quality variables and their relationship in fish farming scenarios.

Dataset files:

training_data.zip: 247 uniformly distributed frames from the entirety of the video footage were annotated for model training and can be found here in various formats including YOLO.

waterquality_data.zip: includes 55 hours and 28 minutes of data from physico-chemical variables collected from a water tank with a Hanna HI98194 probe. The collected parameters were temperature, dissolved oxygen, electrical conductivity, resistivity, total dissolved solids and pH.

behavioral_data.zip: contains the processed variables obtained from the video footage using the trained machine learning model.

weights.zip: the weights used for the YOLOv4 model can be found here.

video_xx.zip: contains 51 hours and 40 minutes of one-minute recordings of the fish tank. Split into 26 two hour sections for download. Useful to extract behavior-related variables.

Version history:

v1: Initial release.

v2: Changed the format for waterquality_data and behavioral_data to fix an error where the timestamp seconds were lost and replaced with 0 when saving and reopening the files.