运动学
帧(网络)
标准差
直线(几何图形)
帧速率
分割
人工智能
计算机视觉
数学
计算机科学
统计
几何学
物理
经典力学
电信
作者
Jialong Zhou,Daxiong Ji,Jian Zhao,Songming Zhu,Zequn Peng,Guoxing Lu,Zhangying Ye
标识
DOI:10.1016/j.compag.2022.107193
摘要
Accurate estimation of fingerlings quantity is one of the most challenging aspects of aquaculture, relevant simple and efficient solutions, however, are still lacking. To address this problem, an accurate and practical on-line counting method for 2–8 cm fingerlings, which was simple to implement, computational inexpensively (without huge pre-training like that in deep learning), and suitable for five different kinds of fingerlings, based on fish kinematics characteristics was proposed in this study. First, the kinematic model of the fingerlings for motion prediction was constructed based on the analysis of their kinematics characteristics; secondly, adaptive threshold segmentation (ATS) algorithm was used to segment and detect the fingerlings; and then, the fingerlings in previous and current frames were associated and tracked by using the kinematic model and the probability density function (PDF). Following this, the new fingerlings were identified and counted in real-time. Finally, through the exhaustive test on 102 datasets (acquired with a low frame rate (10 fps)), the present method showed the average counting accuracy rate (ACAR) and standard deviation (SD) of more than 98.78% and less than 0.95%, respectively. Specifically, the ACAR of Pseudosciaena Crocea (2–3 cm) and Ctenopharyngodon idellus (3–5 cm, 6–8 cm) exceeded 99.19%, and the corresponding SD was less than 0.59%.
科研通智能强力驱动
Strongly Powered by AbleSci AI