The red imported fire ant (solenopsis invicta buren) is one of the major invasive species that has been causing damage in southern China in recent years. Accurately identifying the nests of red imported fire ants is crucial for their prevention and control. In this paper, an improved red fire ant nest identification model based on ResNet34 is proposed to solve the problems of traditional red fire ant prevention and control relying on manual inspection, high risk and low efficiency, so as to realize intelligent inspection. This model uses red imported fire ant nest images collected from different terrain features, combined with data augmentation techniques for training. By adding an SE attention mechanism module after the first convolutional layer of ResNet34 and before the fully connected layer, the model can enhance its adaptive selection and channel weighting adjustment capabilities to extract local nonlinear texture features on the surface of red imported fire ant nests. Through K-fold cross-validation tests and ablation tests to explore hyperparameters, SE-ResNet34 is compared with AlexNet, VGG-16, ResNet18, ResNet34, and ResNet50, and the results show that SE-ResNet34 achieves a peak accuracy of 98.76%, which is 2.17% higher than ResNet34. It has the characteristics of short training time, high recognition accuracy, strong robustness, and stability compared to other tested models. This method provides a convenient and efficient solution for distinguishing red fire ant nests while reducing manual labor and minimizing the use of insecticides.