Introduction
Aquaculture has seen significant changes the recent years, since aquatic resources are considered a primary protein supplement for daily lives in many communities. The aquaculture industry is experiencing a steady increasing production, with an approximate average annual growth rate to 1.72 % since 2010 [1]. Common practices for fish farming require standard procedures that are considered expensive and time-consuming. In recent years, science has presented outstanding novelties, featuring a wide pallette of technological advancements. Frame- and event-based cameras, scanning sonars and light enrichment are some of this wide variety that can create novel methodologies for fish farming.
In our project, two new concepts are introduced: a cyber-enhanced tank and an innovative fish behavior detection via neuromorphic vision [2]. The event camera, which is the basis for the neuromorphic vision, can detect alterations in fish movements by analyzing the generated changes in pixel light intensities. We propose an experimental tank setup with these technological deployments which we believe will provide a more realistic environment for the fish and a multi-modal way of observing fish behavior via a set of (frame and event) cameras and a scanning sonar.
Materials and Methods
Cyber-enhanced tank
The enhanced tank was built upon an existing conventional tank environment of four cubic meters. A waterproof LED line, illuminating the volume homogeneously with a spectrum close to sunlight (5600K to 6500K), was placed a few centimeters below the surface on a specially designed support frame, providing light conditions that mimic those in a commercial cage-mimicking light conditions. As displayed in Figure 1, the lights (operated by a signal frequency driver) follow the inner perimeter of the tank. A surveillance camera (I), a stereo camera (II), an event camera (III) and a sonar (IV), consist a multi-observatory sensor suite.
Neuromorphic vision
A novel event-based fish monitoring procedure is presented via neuromorphic vision, by detecting alterations in pixel brightness (known as events). This differs from the operation of commercial frame-based cameras. Figure 1-III shows an example of outputs from this device, where two highly mobile fish are detected.
The black and blue dots represent the variations in the light intensities of the pixels. The main goal of this procedure was to count the events generated by the fish during different stimuli presentation. As an example, Figure 2 shows feeding operations recorded by the same event camera from two different perspectives, bottom (left) and top (right) placement, respectively. The red dotted lines denote the feeding time while blue lines depict the event time series.
Conclusions
This study introduces two new concepts for aquaculture research: a cyber-enhanced tank that provides a realistic tank environment and a sensor suite for fish monitoring, and a new method for behavioral monitoring based on neuromorphic cameras. The cage-like tank provides light conditions similar to real aquaculture settings, aiming to promote fish welfare during experiments. Neuromorphic vision represents an innovative technology for fish detection, a concept that can provide more information for industrial use.
References
[1]. A. G. J. Tacon, “Contribution of Fish and Seafood to Global Food and Feed Supply: An Analysis of the FAO Food Balance Sheet for 2019,” Rev. Fish. Sci. Aquac., vol. 31, no. 2, pp. 274–283, Apr. 2023, doi: 10.1080/23308249.2022.2124364.
[2]. D. Voskakis et al., “An enhanced and more realistic tank environment setup for the development of new methods for fish behavioral analysis in aquaculture”.