An owl-inspired temporal transformer for enhanced shrimp detection in aquatic environments

This study introduces NOCT-A-VIS, which stands for Nocturnal Adaptive Vision System, symbolizing a bio-inspired framework that mimics owl-like visual adaptations for detecting underwater objects in low-light environments. From the methodological point of view, the physical characteristics of owls' vision from the NOCT-A-VIS framework are incorporated as dedicated computational modules. The proposed shrimp detection transformer analyzes light reflection to emulate the function of the Tapetum Lucidum (TL), which is a deep reflect part of retina exists in owls, improves the visual capability to view the scene and identify the various object under the poor lighting by retinal light trajectory. The TL mechanism focus on the light reflection from retina to the scene or object. The Sensory Enhancement deals with Rod-Inspired Analogous to the high sensitivity of rod cells, the algorithm incorporates a pre-processing step to enhance weak signal detection from underwater sensors must use noise filtering techniques to amplify subtle signals, mimicking the owl's ability to capture minimal light. The visual pigment of the rod cells detects the variations of the object color segments under the water and the captured information transferred to the outer segment of the rod cells. To improve the detection rate of the object a pre-processing step called Empirical Mode Decomposition (EMD) is used to filter the signal noise and adopting the owl visionary technique to identify the objects in the dark mode or in the dim light conditions. The Sequential Process includes Spatial Awareness aspects deals with Large Eye Size-Inspired depends on the biological features of the owl eye and it size is larger than its head so that the perception is good in receiving light to retina for identifying the objects. The complete process working under the curated dataset which is underwater scenes. The proposed pipeline is biologically grounded in three owl-inspired mechanisms. In particular, the sensitivity of rod cells provides insights for weak signal denoising, while a retroreflective role of the tapetum lucidum breeds temporal attention mechanisms in the interest of robustifying low light feature extraction, and finally, binocular visual processing in owls informs the architecture of stereo-depth estimation modules and increases the system's ability for wide-angle spatial awareness.