An adaptive dead fish detection approach using SSD-MobileNet
Guoyan Yu, Lin Wang, Mingxin Hou, Yicha Liang, Taihua He
Abstract
Aiming at the difficulty in the recognition of dead fish in a large cage, a method of detecting dead fish on water surface based on SSD-MobileNet is proposed. In this paper, the models of SSD-MobileNet V1, V2, and V3 are compared, which are suitable for an embedded mobile terminal. Through theoretical and experimental analysis, it has been proved that SSD-MobileNet V3 is not only real-time object detection algorithm, but also a new architecture combining hardware network architecture search (NAS) and NetAdapt. SSD-MobileNet V3 realizes the cooperation between automatic search algorithm and network design, and the complementary method improves the overall technical level.
Topics & Concepts
Computer scienceFish <Actinopterygii>Real-time computingObject detectionArchitectureArtificial intelligencePattern recognition (psychology)FisheryBiologyVisual artsArtWater Quality Monitoring TechnologiesMachine Learning and ELMAdvanced Neural Network Applications