【摘要】 大豆种子由于存储和运输不当，容易产生热损伤问题。热损伤会影响种子的种质质量和发芽率，因此准确地检测热损伤大豆种子对于提高种子品质和农业生产具有重要意义。本文提出了一种基于高光谱成像和多尺度跨通道一维卷积神经网络(MSC1DCNN)的大豆种子热损伤无损检测方法。首先，通过高光谱成像系统获取大豆种子在400～1 000 nm波段的光谱数据，并对比分析不同热损伤大豆种子(正常、轻微热损伤、严重热损伤)的光谱曲线特点。发现在420～500 nm蓝光区域和750～1 000 nm近红外区域，光谱反射率随着热损伤程度的加深逐渐增大。这些变化为后续的热损伤检测提供了有效的光谱特征依据。其次，采用MSC1DCNN模型进行分类，该模型在测试集上的准确率、召回率和F1分数均达到99.07%,优于支持向量机(SVC)(F1分数为88.32%)、 k-近邻算法(KNN)(F1分数为84.39%)及一维卷积神经网络(1D CNN)(F1分数为92.90%)。特别地，MSC1DCNN模型在鉴别轻微热损伤与正常大豆种子时误判率为1.39%,显著低于SVC(12.04%)、 KNN(15.74%)和1D CNN(9.72%)模型。最后，还通过发芽试验验证了热损伤对大豆种子发芽率的影响。实验结果表明，热损伤显著降低了大豆种子的发芽率，进一步证实了热损伤对大豆生长的潜在危害。综上所述，本研究提出的MSC1DCNN模型为热损伤大豆种子的无损检测提供了一种有效解决方案，对种质质量检测和自动化筛选工作提供了新的思路。更多还原

【Abstract】 Soybean seeds are prone to heat damage due to improper storage and transportation. Heat damage affects the seed quality and germination rate, making it crucial to accurately detect heat-damaged soybean seeds for improving seed quality and agricultural production. This paper proposes a non-destructive detection method for heat-damaged soybean seeds based on hyperspectral imaging and a Multi-scale Cross-channel One-dimensional Convolutional Neural Network(MSC1DCNN). Firstly, hyperspectral imaging systems were used to capture spectral data of soybean seeds in the 400～1 000 nm wavelength range. The spectral curves of different heat-damaged soybean seeds(normal, mild heat damage, and severe heat damage) were compared and analyzed. It was found that the spectral reflectance in the 420～500 nm blue light region and the 750～1 000 nm near-infrared region gradually increased with the degree of heat damage. These spectral variations provided effective spectral features for subsequent heat damage detection. Secondly, the MSC1DCNN model was applied for classification. The model achieved an accuracy, recall, and F1 score of 99.07% on the test set, outperforming Support Vector Classification(SVC)(F1 score of 88.32%), k-Nearest Neighbor(KNN)(F1 score of 84.39%), and One-dimensional Convolutional Neural Network(1D CNN)(F1 score of 92.90%). Notably, the MSC1DCNN model had a misclassification rate of 1.39% in distinguishing mild heat-damaged seeds from normal seeds, which was significantly lower than SVC(12.04%), KNN(15.74%), and 1D CNN(9.72%). Finally, a germination experiment was conducted to verify the effect of heat damage on the germination rate of soybean seeds. The experimental results demonstrated that heat damage significantly reduced the germination rate of soybean seeds, further confirming the potential harm of heat damage to soybean growth. In conclusion, the MSC1DCNN model proposed in this study offers an effective solution for the non-destructive detection of heat-damaged soybean seeds, providing new insights for seed quality detection and automated screening.更多还原