Intent classification in question answering systems is vital for providing accurate and relevant answers. In the area of apple pest and disease inquiries, challenges are presented by semantic complexity and word-order dependency. To tackle these obstacles, we proposed an Attention-enhanced Bidirectional Gated Recurrent Unit and Multi-level Convolutional Neural Network (Att-BiGRU-MulCNN) model. First, the Bidirectional Gated Recurrent Unit (BiGRU) is utilized to capture the temporal dependencies in farmers’ queries. It models the semantic relationships within these queries. Concurrently, an attention mechanism is integrated into the BiGRU, which is pivotal in highlighting the key features, ensuring that the model focuses on the most relevant information. Subsequently, at the classification layer, a Multi-level Convolutional Neural Network (MulCNN) is incorporated. The MulCNN capitalizes on multi-level convolutional kernels to extract local semantic features from the queries. This extraction process bolsters the model’s discriminative ability, enabling it to efficiently distinguish between diverse intent categories. Finally, the proposed Att-BiGRU-MulCNN model is evaluated on a self-constructed dataset. This dataset comprises 47,762 apple-related queries, which were manually labeled into nine predefined intent categories. The experiment results demonstrate that the proposed approach outperforms five benchmark intent classification models in terms of accuracy, recall, and F1 score. The Att-BiGRU-MulCNN model accomplishes precise intent recognition in the domain of apple pest and disease inquiries, thereby providing a reliable solution to this problem.