FPGA-Based Real-Time Embedded Fish Embryo Detection System

Abstract

In today’s aquaculture industry, the period of fish embryos needs to be detected in real time because different periods of fish embryos require different environments for their cultivation. The paper proposes an FPGA-based real-time embedded fish embryo detection system to solve the problems of today’s fish embryo period detection, which consumes a lot of human resources and has low accuracy. Based on the selection of YOLOv3 as the basic model, this paper combines model optimization and hardware acceleration to make the whole system achieve the effect of real-time detection. With only a 4.23% reduction in the mAP value of the model, the inference speed of the model was improved by nearly 150%. Finally, we achieved 34.1 fps of real-time fish embryo detection on a 100 MHz FPGA chip, while having 82.49% mAP, and the whole embedded system only consumed 2.524w while running. The results show that the FPGA-based real-time fish embryo detection system proposed in this paper is fully capable of meeting the needs of today’s aquaculture embedded devices. It is beneficial to promote the transformation of fish embryo analysis from traditional manual methods to AI-based embedded devices and facilitate the rapid analysis of fish embryos.

Lifelong Learning Architecture of Video Surveillance System

“…A particular size was applied to the inputs for those benchmark techniques, and training was done with the default options utilized in each algorithm's official, publicly available code. We adopted SSD [16] as a baseline backbone in the case of MobileNet v3 [17] and PeleeNet [18]. K-means clustering was used to obtain the anchor size for each training data sets.…”

Privacy-preserving AI-enabled video surveillance for social distancing: responsible design and deployment for public spaces

“…One-stage models are designed to improve the inference time of two-stage models while still achieving comparable accuracy (Liu et al , 2016; Redmon and Farhadi, 2018). Mobile-based models are the compact version of the previous models dedicated to specific applications on smaller computing devices such as mobile and edge devices (Wang et al , 2018). Mobile-based models are built on one-stage model with smaller neural networks for faster and cheaper computation (Biswas et al , 2019).…”

College students’ perceptions of AI-based writing learning tools: With a focus on Google Translate, Naver Papago, and Grammarly

“…Building on this framework, some variants were proposed. PeleeNet [79] successfully proposed modifications to achieve real-time inference capabilities upon DenseNet. VovNet [40] departed from DenseNets' dense feature reuse in favor of a sparser one-shot aggregation aimed at real-time object detection.…”

“…Those architectures surpassed ViTs on ImageNet and dense prediction tasks as well, but similarities like using additive shortcuts and architectural complexity continue to restrict architectural diversity and innovation. Furthermore, network modernization methods [5,48,79,84] have successfully revisited existing architectures but did not handle beyond baselines using additive shortcuts. This work follows a general direction but ensures our starts from a distinct baseline, acknowledging uncertainties about the effectiveness of existing roadmaps.…”