Deep Learning in Rice Pest Recognition, to address the problems of large number of parameters and high requirements on training samples and arithmetic power of traditional CNN models, this paper proposes a lightweight L-V Net model based on VovNet. The model replaces the ordinary convolution in VovNet with depth-separable convolution to reduce the model parameters and improve GPU utilization. In addition, a normalized channel attention mechanism is introduced at the end of the model to enhance the feature extraction capability and control the number of network parameters. The study used 5,785 RGB images containing 12 types of common rice insect pests such as blind stink bugs, locusts, and red spiders as test data and achieved 98.14% recognition accuracy. The L-V Net model has fewer parameters, lower complexity, lower network latency, and higher recognition accuracy.