ShrimpFarming Dataset

Description

 

 

About:

The ShrimpFarming dataset documents the behavior of shrimp around a feeding plate in an aquaculture tank, to be used for extracting various indicators about the monitored shrimp and their behavior.

The images from the dataset were acquired at RiaSearch, Aveiro, Portugal, on five different days, at different times of the day and under varying illumination conditions. The image acquisition was manually triggered by pressing a button at the moment the feeding pellets were introduced in the tank. A Raspberry Pi Cam Module 3 WIDE was used for image acquisition, which was positioned approximately 95 cm above the surface of the tank.

The ShrimpFarming dataset captures shrimp interactions across two distinct conditions: (i) Feed, which includes images taken during active feeding periods when pellets are present in the tank, and (ii) NoFeed, which captures shrimp behavior outside of feeding times, with no pellets in the feeding plate.

The dataset contains one capture of the “NoFeed” condition and six captures during feeding, taken in a span of five different days. Three captures occurred around 11 a.m. in both sunny and cloudy conditions, two around 2 p.m., and one at 5 p.m., experiencing a mix of sun and clouds. Among these six captures, three use pellets of type A and the remaining three use pellets of type B, which vary in both color and shape.

The ShrimpFarming dataset includes two types of annotations: (i) anatomy, which differentiates between shrimp based on their positioning and orientation, allowing for more detailed analysis of shrimp posture and shrimp measurements, and (ii) activity, which provides general annotations for all shrimp without distinguishing individual positions. In both cases annotations for the feeding plate position are included.

The duration of each capture session was 20 minutes long, but two distinct timing protocols were followed: (i) images were captured every 4 seconds; and (ii) the captures follow a structured image acquisition protocol designed to emphasize the initial moments of feeding activity. In the latter case, images were captured every 4 seconds for the first 2 minutes, every 8 seconds for the next 2 minutes, and then every minute for the remaining 16 minutes, offering a detailed view of shrimp behavior throughout the feeding period.

 

ShrimpFarming Dataset Organization:

·       The root folder, named "ShrimpFarming dataset", contains two subfolders: (i) "Feed", corresponding to the captures that occurred during feeding (include pellets); (ii) “NoFeed”, which includes the captures that did not happen in feeding period (do not include pellets).

·       These two subfolders contain subfolders called “type_X”, where "type" corresponds to if the capture is “Feed” or “NoFeed” and "X" is the number (identifier) of the capture.

·       Inside each “type_X” folder are three subfolders and a metadata file: (i) “images”, which contains the images of the corresponding “typeX” capture; (ii) “activity_labels”, which contains all the activity annotations of those images; (iii) “anatomy_labels”, that has the anatomy annotations of the images; and (iv) “metadata_typeX”, which is a .yaml file that contains information regarding the “type_X” capture.

·       Inside the “images” folder are the images of the corresponding capture that follow the naming convention: “typeX_IMGYYYY”, where “typeX” regards the capture type and the capture identifier, and “YYYY” is the sequential number of the image from the capture "X".

·       The “activity_labels” and “anatomy_labels” folders contain the annotations of the images in YOLO format. Each image has a corresponding .txt annotation file, in each of these two folders, whose names are the same as the images: “typeX_IMGYYYY”.

·       The “metadata_typeX.yaml” files provide specific information about each capture session, such as date, time, capture duration, capture intervals protocol, number of shrimps in the tank, water temperature, shrimp breed and the pellets used (A or B).

 

Highlights:

·       The IST EURECOM Light Field Face Database is the first database to include the raw light field images, sample 2D rendered images and the corresponding depth maps along with a rich collection of metadata, including the location of a set of facial landmarks.

 

License agreement:

·       To use the database please fill in the license agreement and send a scanned copy of the signed form by e-mail.

 

Download:

·       The database can be downloaded from the link below.

·       A password for decrypting the compressed Zip file will be provided after receiving the duly signed license agreement (see above).

part01

 

How to reference the database:

·       B. Correia; O. Pacheco; R.J.M. Rocha; P.L. Correia; “Image-Based Shrimp Aquaculture Monitoring”. Sensors 2025, Vol. 25, No. 248. https://doi.org/10.3390/s25010248

 

Contacts:

Feedback is welcome. Please provide your comments and suggestions by e-mail.