Large-scale jellyfish outbreaks have caused a severe threat to both human life and marine ecology. Therefore, jellyfish-detecting technology has garnered a lot of interest. The paper investigates jellyfish detection and classification algorithms based on optical imagery and deep learning theory. First, an underwater image enhancement algorithm is proposed. In addition, the article creates a dataset of 11926 photos that contains seven jellyfish species and fish. An improved YOLOv4-tiny algorithm is suggested based on the Convolutional Block Attention Module and a better training approach. According to the results, the accuracy of the improved algorithm reaches 95.01%, which is 1.55% higher than the YOLOv4 algorithm and 2.55% higher than the YOLOv4-tiny algorithm. Additionally, the detection speed is 223 FPS, substantially faster than the YOLOv4 algorithm's 43.9 FPS. In conclusion, our method can detect the jellyfish accurately and quickly. The paper establishes the groundwork for developing a real-time submarine jellyfish monitoring system.
This paper builds a real-time infrared target detection system with a visual positioning function based on FPGA, a visual camera, and an infrared camera. Firstly, the whole system is built based on FPGA. The design and implementation of the infrared image acquisition module, a data storage module, SD card module, and image display module are completed. Ping pong operation is used for the data storage modules to realize video stream transmission. The experiment results show that the system built in this paper can collect and display the target in real-time. To improve the system performance, several image processing algorithms are proposed, including an improved median filtering algorithm, linear transformation, and Laplacian sharpening algorithm, a combined algorithm of histogram equalization, Gamma transform, Laplacian sharpening, target detection algorithm combined with threshold segmentation and background difference algorithm, and visual localization algorithm. Software simulation and FPGA hardware implementation results show the effectiveness of the proposed algorithms.
In order to solve the problem that the gridless DOA estimation algorithms based on generalized finite rate of innovation (FRI) signal reconstruction model are not suitable for two-dimensional DOA estimation using planar array, a separable gridless DOA estimation algorithm exploiting bi-orthogonal sparse linear array (BSLA) structure is proposed in this paper, which is called 2D-SGFRI. The 2D-SGFRI algorithm firstly recovers the covariance data of the virtual array formed by BSLA through the matrix completion method, so as to obtain the complete covariance data vectors about two independent parameters respectively. Next, since the covariance data vector satisfies the constraints of annihilation filter equations, the generalized FRI signal reconstruction model can be utilized to retrieve DOA from the covariance data vector. Compared with the existing DOA estimation algorithms based on generalized FRI signal reconstruction model, the 2D-SGFRI algorithm can be can be effectively applied to twodimensional DOA estimation, and can obtain stable estimation results. At the same time, due to the reduction of the dimension of positive semidefinite matrix, the 2D-SGFRI algorithm can significantly reduce the computational complexity compared with the two-dimensional DOA estimation algorithms based on atomic norm minimization (ANM). A series of simulation experiments are shown to verify the effectiveness and superiority of 2D-SGFRI algorithm. INDEX TERMS Two-dimensional DOA estimation, bi-orthogonal sparse linear array (BSLA), finite rate of innovation (FRI), matrix completion, annihilation filter
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.