Tea Growth Monitoring and Intelligent Retrieval Based on the RAM-BLIP Model

In response to the practical demands of insufficient cross-modal retrieval accuracy and scarce domain data in the tea plant growth monitoring scenario, this paper proposes a tea garden image-text retrieval algorithm based on an improved BLIP framework. With BLIP as the backbone, the RAM spatial attention mechanism is integrated into the visual encoder to highlight local significant regions and suppress irrelevant textures, significantly enhancing the model's perception of fine-grained growth features. A semantic transformation layer is added to the text segment to explicitly model domain keywords such as variety names and growth stages. In the cross-modal interaction stage, gated bidirectional attention is introduced to adaptively allocate weights between the image-to-text and text-to-image information flows, suppressing redundant noise while improving alignment quality. Experimental results show that this method outperforms other mainstream models on multiple public datasets, verifying its efficiency and robustness in image-text matching tasks.