| 摘要: |
| 为探索应用计算机听觉技术实现对工厂化循环水养殖鳗鲡数量的评估,建立了一种基于回归分析的SENet网络模型。针对数据集中包含的白噪声声谱图数据缺乏可利用的动态规律问题,通过修改SENet输出层、输出范围、评价指标等,使其直接利用图像进行回归分析任务,从而进一步提高了网络在图像分析任务上的性能。在循环水养殖鳗鲡的数量评估试验中,设置8组不同的鳗鲡数量进行试验,结果表明：水听器接收到的声音信号与鱼数量呈现出明显的相关性；在测试阶段SENet网络的拟合相关系数为0.98,SENet回归分析模型在测试集样本上的决定系数R2、平均绝对误差MAE和平均绝对百分比误差MAPE分别为0.96、1.66和4.43%；采集了30组数据样本对训练好的模型进行验证试验,SENet模型预测数量的相对误差与变异系数都在8%以内,预测准确率达到90%以上。 |
| 关键词: 计算机听觉技术 鳗鲡 鱼群数量评估 声音信号 SENet网络 |
| DOI： |
| 分类号: |
| 基金项目:国家重点研发计划, 2020YFD0900102号; 鳗鲡现代产业技术教育部工程研究中心开放基金, RE202304号, RE202101号。 |
|
| Evaluation of eel (Anguilla) number in industrial recirculating aquaculture based on SENet |
|
LIN Xi1,2, JIANG Xing-Long1,2, ZHOU Shi-Hao1,2
|
|
1. Fisheries College, Jimei University, Xiamen 361021, China;2. Engineering Research Center of the Modern Technology for Eel Industry, Ministry of Education, Xiamen 361021, China
|
| Abstract: |
| A SENet network model based on regression analysis was established to evaluate the number of eel cultured in the recirculating aquaculture system (RAS). Given the lack of available dynamic laws of the white noise spectrograph data included in the data set, by modifying SENet output layer, output range, evaluation index, etc., SENET directly used images for regression analysis tasks, thus further improving the performance of the network in image analysis tasks. In the evaluation test of eel number in the RAS, 8 groups of different eel number were set up. The results showed that there was obvious correlation between the acoustic signal received by hydrophone and fish number. The fitting correlation coefficient of SENet network was 0.98 during the test stage. The determination coefficient R2, mean absolute error MAE and mean absolute percentage error MAPE of SENet regression analysis model on test set samples were 0.96, 1.66 and 4.43%, respectively. Thirty sets of data samples were collected for validation tests on the trained model. The relative error and coefficient of variation of SENet model for predicting eel population were both less than 8%, and the prediction accuracy was over 90%. |
| Key words: computer auditory technology eel assessment of fish number sound signal SENet network |