基于深度学习的鳝鱼新鲜度评估方法

Evaluation method of the freshness of Monopteros albus based on deep learning

中文关键词:  深度学习  机器视觉  卷积神经网络  鳝鱼  新鲜度  分类网络

英文关键词:deep learning  machine vision  convolutional neural networks  Monopteros albus  freshness  classified networks

中文摘要:

      目的 构建基于深度学习的鳝鱼新鲜度评估模型。方法 本研究首先根据自动凯氏定氮法测定鳝鱼肉的挥发性盐基氮(total volatile base nitrogen, TVB-N), 构建了鳝鱼新鲜度分级图像数据库。选择You Only Look Once version 5 (YOLOv5)、You Only Look Once version 8 (YOLOv8)以及Regions with CNN features (R-CNN) 3种神经网络建立检测模型, 最后, 针对3种神经网络选择准确率、平衡F分数、平均精度均值3个指标来评估模型的性能。结果 根据YOLOv5、YOLOv8、R-CNN的鳝鱼新鲜度识别准确率分别为97.11%、83.94%、91.67%, 最终选择效果较优的YOLOv5作为分类算法模型。结论 本研究建立的方法操作简单, 高效方便, 有助于保障食品安全和提高评估效率, 推动了智能化水产养殖和加工的发展。

英文摘要:

      Objective To construct a deep learning-based Monopteros albus freshness evaluation model. Methods In this study, a database of Monopteros albus freshness classification images was constructed by determining the total volatile base nitrogen (TVB-N) of Monopteros albus meat based on the automatic Kjeldahl nitrogen determination method. Three kinds of neural networks, You Only Look Once version 5 (YOLOv5), You Only Look Once version 8 (YOLOv8) and Regions with CNN features (R-CNN), were selected to establish the detection model, and finally, 3 kinds of indicators of the accuracy, balanced F score and mean average precision were selected for 3 neural networks to evaluate the performance of the model. Results According to YOLOv5, YOLOv8 and R-CNN, the accuracy of Monopteros albus freshness recognition was 97.11%, 83.94% and 91.67%, respectively, and YOLOv5 with the best effect was selected as the classification algorithm model. Conclusion The method established in this study is simple, efficient and convenient, which is helpful to ensure food safety and improve evaluation efficiency, and promote the development of intelligent aquaculture and processing.