基于深度学习的番茄叶部病害识别方法研究

Research on Tomato Leaf Disease Recognition method based on Deep Learning

在深度学习中,番茄叶部病害图像识别模型往往采用传统的卷积神经网络,虽然识别率高、性能好,但存在参数量大、成本高、训练时间长等问题,在硬件资源受限的环境条件下局限性较大,难以部署到移动设备、嵌入式设备等终端。基于此,在ResNet18的主体结构上,通过增加特征提取尺度、更新残差层连接方式、分解大卷积核等操作,采用了一种改进型的病害识别模型——Multi-scale ResNet。在减少参数量的同时降低了空间存储开销。试验结果表明,与ResNet18相比,在准确率相差不大的情况下,模型的训练参数减少约99%。提出的改进型网络在保证精确度的前提下降低了模型的复杂度,使番茄叶部病害识别模型在硬件资源受限的条件下仍可以运行部署,更具有普适性。

In deep learning,several traditional convolutional neural networks are often used in tomato leaf disease image recognition models.Although the recognition rate is high and the performance is good,there are still problems such as large parameter quantities,high cost,and long training time.In an environment with limited hardware resources,there are still significant limitations and it is difficult to deploy to mobile devices,embedded devices,and other terminals.Based on the main structure of ResNet18,adopted an improved disease identification model,Multiscale ResNet,by adding feature extraction scales,updating residual layer connectivity,and decomposing large convolution cores.Reducing the amount of parameters while reducing the space storage overhead.The experimental results showed that compared with ResNet18,the training parameters of the model were reduced by about 99% with little difference in accuracy.The improved network proposed in this study reduced the complexity of the model while ensuring accuracy,making the tomato leaf disease identification model still operational and deployable under limited hardware resources,making it more universal.