.png){kind=icon}
.png){kind=icon}
Return
Group activity amount estimation for fish using multi-object tracking
.png){kind=icon}
PRE
AI
DOI：10.1016/j.aquaeng.2025.102534.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The activity amount of fish can directly indicate its level of health. Quantitative indexes of fish activities provide valuable data for water quality control, disease warning, and fish behavior research, which can enhance factory breeding technology and fish welfare. To address the issue of a lack of indicators for the amount of activity in fish groups and individuals, a group activity amount estimation method based on fish multi-object tracking is proposed in this paper. The FasterNet network is improved by incorporating a branch that learns the features from the previous frame. This method is referred to as Improved FasterNet Fish Tracking (IFFT), and is capable of fast and accurate identification, location, and estimation of the offset of fishes between the two frames in a joint way. Subsequently, a method of activity amount estimation based on grid step is proposed to get the activity index from the trajectories. Moreover, a visualization method is proposed to illustrate the spatial distribution of fish groups. This method utilizes elliptic fitting to calculate the spatial distribution characteristics such as radius, dispersion, cohesion, and polarization. Experiments conducted on the established fish tracking dataset demonstrate that the proposed IFFT achieved a high-order tracking accuracy (HOTA) of 67.54 % and a multi-object tracking accuracy (MOTA) of 92.4 %, surpassing other existing mainstream methods. Furthermore, the average tracking speed reached 20.6 frames per second (FPS). The mean absolute percentage error (MAPE) of the activity amount estimation of fish groups based on the grid step is 0.15 % per frame, indicating a highly accurate estimation of fish activity. The results demonstrate that the proposed activity estimation method of fish groups based on multi-object tracking effectively eliminates the interference caused by water movement. It also achieves a balance between speed and accuracy, enabling real-time quantitative and visual analysis of fish group movement and providing decision support for improving the factory farming process.
Keywords:
Multi-object tracking
Fish
Activity amount estimation
Group tracking
Improved FasterNet
Visualization analysis
Journal
IF:
4.3
Papers:
1.6K
Citations:
4.2K