基于深度学习的农业病害图像识别的研究

Research on Image Recognition of Agricultural Diseases Based on Deep Learning

随着深度学习技术在图像识别任务中的不断发展与成熟,其算法逐渐被应用于农业病虫害的识别中,并取得较好的效果。但是,农业病害图像的识别比农业病虫害的识别更具有挑战性。因此,将ResNet、BotNet、CotNet作为Baseline,首先,通过引入通道注意力机制、MHSA模块和Contextual Transformer Networks模块相结合以改进网络模型结构,使得改进后的网络模型可以更好地提取图像的全局特征信息和局部特征信息,提高网络模型的特征表达能力;其次,将正交投影损失函数与传统交叉熵损失函数相结合,降低图像分类网络模型在农业病害数据集上在训练过程中标签噪音等不利因素的干扰,以达到优化农业病害识别训练过程和结果。最终通过多组实验证明,Baseline模型的改进与训练过程的优化,可以有效提高模型在农业病害数据集上的分类的准确率和鲁棒性并且网络模型参数量和计算量减少,使得改进和优化后的农业病害识别模型可以更适用于实际的农作物病害识别工作中去,为农作物病害处理助力、为智慧农业发展赋能。

With the continuous development and maturity of deep learning technology in image recognition task,its algorithm is gradually applied to the recognition of agricultural pests and diseases,and achieves good results.However,the recognition of agri-cultural disease images is more challenging than that of agricultural diseases and pests.Therefore,taking RESNET,botnet and cot-net as the baseline,firstly,the structure of the network model is improved by introducing the channel attention mechanism,MHSA module and contextual transformer networks module,so that the improved network model can better extract the global and local fea-ture information of the image,and improve the feature expression ability of the network model.Secondly,the orthogonal projection loss function is combined with the traditional cross entropy loss function to reduce the interference of adverse factors such as label noise in the training process of image classification network model on agricultural disease data set,so as to optimize the training pro-cess and results of agricultural disease recognition.Finally,through several groups of experiments,it is proved that the improve-ment of the baseline model and the optimization of the training process can effectively improve the accuracy and robustness of the classification of the model on the agricultural disease data set,and the amount of parameters and calculation of the network model is reduced,so that the improved and optimized agricultural disease identification model can be more suitable for the actual work of crop disease identification,help the treatment of crop diseases,and empower the development of intelligent agriculture.