To address the problems of feature redundancy and insufficient utilization of hierarchical features in the classification process of convolutional neural network models in deep learning, this paper proposes a novel deep learning model termed as Gated Convolution and Feature Pyramid Network (GCFPN). This model incorporates a gating mechanism in the conventional convolutional layers to filter redundant features and extract useful temporal features, and then utilizes a feature pyramid structure to integrate features from different hierarchical levels, thereby obtaining more comprehensive feature information for input into the classification module. Furthermore, the model is trained using the Focal Loss function to enhance its attention towards minority class and hard-to-classify samples during training process. The experimental results show that the proposed GCFPN model outperforms the comparison algorithms in various detection indicators, improving the weighted average of Precision, Recall, and F1 scores of all samples in the test set by 3.27%, 3.42%, and 6.13%, respectively.