For individual radiation source identification technology using deep neural networks, network depth is continuously increased to achieve good recognition performance, resulting in an explosion of model parameters and computational complexity, which makes deployment difficult on resource-constrained edge devices. To address this, this paper proposes a network architecture called ODCNet(One-Dimensional Depthwise Separable Convolution Network) based on one-dimensional depthwise separable convolution and one-dimensional convolutional block attention modules. By combining depthwise and pointwise convolutions, one-dimensional depthwise separable convolution effectively reduces model parameters and computational complexity. The lightweight one-dimensional convolutional block attention module can effectively enhance model performance and ensure recognition capability. Experimental results show that ODCNet's recognition performance is comparable to MobileNet V3, while its parameters are only 11.27% of MobileNet V3's, its computational complexity is 17.49% of MobileNet V3's, and its inference time is reduced to 50% of MobileNet V3's.