基于卷积神经网络的家蚕农药中毒识别研究

Research on Recognition for Pesticide Poisoning of Silkworm Based on Convolutional Neural Network

由农药污染桑叶造成的家蚕中毒是蚕茧生产减收的主要原因之一。现有的家蚕农药中毒识别方法主要是人工观察大量蚕儿出现的明显症状进行识别。为了提高诊断的效率和准确性,运用深度学习方法的卷积神经网络开展了家蚕农药中毒的智能识别研究。通过给4龄初期幼虫添毒的方式集中获取敌杀死(溴氰菊酯)、草甘膦2种农药中毒的家蚕样本,并以相同日龄的健康家蚕作为参照,使用智能手机在实际环境下采集图像,构建家蚕中毒症状的图像数据集。在此基础上,利用轻量级卷积神经网络MobileNet模型开展识别试验,结果显示对2种农药中毒的识别准确率分别为95.0%和99.6%。初步的研究结果提示,进一步构建种类更加完整的基于卷积神经网络的家蚕农药中毒识别系统,将有助于蚕桑生产上家蚕中毒的农药污染溯源分析和精准预防。

Silkworm poisoning is one of the main reasons for the reduction of silkworm cocoons in China. Accurate recognition of poisoning types is helpful for traceability analysis and precise control. The existing method for poisoning recognition is mainly manual recognition, which can only be effective under the conditions of there are a large number of poisoned silkworms. To this end, the method of deep learning was used to recognize silkworm poisoning in this paper. The silkworm samples suffered from two pesticides were collected by feeding and infection. A smartphone was used to collect images in the actual environment, the healthy silkworms were also utilized for recognition, and a dataset of silkworm poisoning images was constructed. The recognition experiment was carried out by using the lightweight convolutional neural network MobileNet, and the results showed that the recognition accuracy of the two types of poisoning was 95.0% and 99.6%, respectively. This study shows that the convolutional neural network can efficiently and accurately identify the types of silkworm poisoning, which can provide a reference for poisoning traceability analysis and precise prevention and control research.