基于拉曼光谱和深度学习的家蚕卵微粒子病无损检测

Non-Destructive Detection of Silkworm Pebrine Disease at Egg Stage Based on Raman Spectroscopy and Deep Learning

为研究家蚕微粒子病检测方法,基于密集连接块提出用R-DenseNet模型对家蚕微粒子病拉曼光谱进行无损检测。以患家蚕微粒子病原原母种卵为实验样本,构建家蚕微粒子病拉曼光谱数据集。R-DenseNet与其他5种分类模型的对比结果表明,不使用额外预处理的R-DenseNet的检测准确率达到97.32%,优于使用预处理的传统分类模型;对于处理60 dB强度噪声的光谱数据,R-DenseNet能达到93.66%的检测精度,在同等性能中,其模型训练的参数量较对比模型减少50%以上,表现出更好的鲁棒性和计算效率。文中提出的R-DenseNet网络结构能够对家蚕卵微粒子病拉曼光谱实现快速、准确且无损的检测,为家蚕微粒子病检测提供了一种新途径。

In order to study the detection method of silkworm pebrine disease, we proposed R-DenseNet based on dense connected blocks for non-destructive detection of silkworm pebrine disease in Raman spectra. R-DenseNet was compared with five other classification models, and the results showed that the detection accuracy of R-DenseNet without additional preprocessing reached 97.32%, which was better than the traditional classification models with preprocessing. R-DenseNet achieve 93.66% for spectral data processing 60 dB intensity noise, and its model training parametric number reduced by more than 50% compared with the comparison model in the same performance, showing better robustness and computational efficiency. The proposed R-DenseNet network structure can achieve fast, accurate and non-destructive detection of silkworm egg pebrine disease Raman spectra, providing a new way for silkworm pebrine disease detection.