%0 Journal Article %T 基于触压反力与深度学习的盐渍海参熟度识别
Cooking Degree Recognition of Salted Sea Cucumber Based on Contact Pressure Reaction and Deep Learning %A 侯宇佩 %A 张旭 %A 崔佳鹏 %A 隋倩倩 %A 王显强 %A 孙纬宸 %J Software Engineering and Applications %P 285-295 %@ 2325-2278 %D 2026 %I Hans Publishing %R 10.12677/sea.2026.152027 %X 针对盐渍海参煮制过程中熟化程度人工触压检测主观性强、效率低、难以量化的问题，本研究旨在构建基于触压力学特征与深度学习的海参熟度精准识别方法。以大连仿刺参盐渍原料为研究对象，依据感官评定将其分为对照组(未煮制)、三成熟、五成熟、七成熟、全熟及过熟6个熟度等级，通过质构仪模拟人工触压开展整参压缩试验，提取21个位移点对应的触压力–位移时序特征数据；选取KNN、随机森林、SVM三种传统机器学习模型与1DCNN深度学习模型进行对比，引入果蝇优化算法对1DCNN模型超参数进行智能寻优，提升模型熟度识别性能。结果表明，1DCNN深度学习模型适配触压力–位移时序特征提取需求，基础模型整体识别准确率达95.83%，其中对照组、三成熟、七成熟样本识别准确率为100%；经果蝇优化算法优化后，模型训练初始准确率从20%提升至80%，最终整体识别准确率提升至98.75%，较优化前提升2.92%，损失值从0.22降至0.10；优化后对照组、三成熟、五成熟、七成熟样本识别准确率均达100%，过熟样本识别准确率从85%提升至97.5%，有效解决了相邻熟度误判问题。本研究证实基于触压力–位移时序特征与果蝇优化1DCNN模型的识别方法，可实现盐渍海参6个熟度等级的精准、高效识别，为开发海参加工熟度快速、客观、无损的自动化检测技术提供了理论依据与数据支撑。
Aiming at the problems of strong subjectivity, low efficiency and difficulty in quantifying the cooking degree of salted sea cucumber during cooking, this study aims to build a precise identification method of sea cucumber cooking degree based on the characteristics of touch mechanics and deep learning. Taking the salted raw material of Apostichopus japonicus in Dalian as the research object, according to the sensory evaluation, it was divided into six cooking degree levels: control group (uncooked), three cooking degree, five cooking degree, seven cooking degree, full cooking degree and over cooking degree. The integrated parameter compression test was carried out by using texture analyzer to simulate the artificial contact pressure, and the characteristic data of contact pressure displacement time series corresponding to 21 displacement points were extracted; Three traditional machine learning models, namely KNN, random forest and SVM, are selected to compare with 1DCNN deep learning model. The fruit fly optimization algorithm is introduced to intelligently optimize the 1DCNN model’s super parameters to improve the model’s cooking degree recognition performance. The results show that the 1DCNN deep learning model adapts to the needs of feature extraction of contact pressure displacement time series. The overall recognition accuracy of the basic model is 95.83%, and the recognition accuracy of the control group, three mature and seven mature samples is 100%; After optimization by Drosophila optimization algorithm, the initial accuracy of model training increased from 20% to 80%, and the final overall recognition accuracy increased to 98.75%, increased by 2.92% compared with that before optimization, and the loss value decreased from 0.22 to 0.10; After optimization, the recognition accuracy of the control group, three %K 盐渍海参， %K 熟度识别， %K 1DCNN， %K 深度学习， %K 果蝇优化算法
Salted Sea Cucumber %K Cooking Degree Recognition %K 1DCNN %K Deep Learning %K Fruit Fly Optimization Algorithm %U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=140855