Scene Recognition with Limited Capture using Domain Adaptation

Scene recognition plays a crucial role in various computer vision applications. However, current methods often face the challenge of incomplete input samples, where only partial or local regions of a scene are available for recognition. To address this issue, we propose a scene recognition method based on multi-source target sample adaptive learning, where the source samples capture the global representation of scene images allowing for alleviate this issue. Our approach leverages data collected from other source samples to learn scene recognition knowledge, which is then used to guide the training of the current model through features learned with an attention mechanism. Additionally, we design a loss function based on multi-source adaptive learning to further enhance the model's performance. Our method learns representative feature of local scene guided by multi-source scene dataset based on attention model, and the domain shift is alleviate by domain adaptive objective for common feature learning. Experimental results demonstrate that our method outperforms existing approaches in terms of recognition accuracy and cross-domain adaptability.