A Lightweight Crop Pest Classification Model based on VovNet
Due to the large number of parameters of the traditional CNN model and the high requirements for training samples and computing power,the use of deep learning for crop pest identification is likely to cause crop identification to be limited by hardware conditions.In this paper,a lightweight LVovNet model is designed based on VovNet.The ordinary convolution of VovNet is replaced by deep separable convolution,which reduces the model parameters and improves the GPU utilization.At the end of the model,the normalized channel attention mechanism is added to strengthen the network feature extraction ability and control the number of network parameters.A total of 5 785 RGB images of 12 common crop pests categories such as mirids,locusts and red spiders were used as test data.The recognition accuracy of the model was 97.34%.Compared with VGG,ResNet,DenseNet and VovNet,it has the characteristics of less parameters,low complexity,low network delay and high recognition accuracy.
pest classificationCNNdepth-separable convolutionchannel attention
NETL
NSTL
万方数据
