NIGHT SHIPPING IDENTIFICATION BASED ON RESIDUAL NETWORK AND ATTENTION MECHANISM
-
摘要: 针对常规的深度学习模型对夜间监控视频进行识别的效果不佳，提出一种基于残差网络和注意力机制的夜间航运事件的识别方法。增强夜间监控视频生成的暗光图像中的光照，采用SE-R2(2+1)模型对增强图像组合成的视频进行识别。该识别模型基于R(2+1)D模型，通过改进模型的激活结构，提升模型的泛化能力。嵌合SENet网络来提高模型的表征能力。实验结果表明，在增强后形成的数据集下，该方法识别准确率达到了88.2%，验证了模型的有效性。Abstract: Aimed at the poor performance of conventional deep learning models in identifying night surveillance videos, a method for identifying night shipping events based on residual network and attention mechanism is proposed. The illumination in the dark image generated by the night surveillance video was enhanced, and the SE-R2 (2+1) model was used to identify the video combined by the enhanced image. The recognition model was based on the R(2+1)D model. By improving the activation structure of the model, the generalization ability of the model was improved. At the same time, the SENet network was embedded to improve the characterization ability of the model. Experimental results show that under the enhanced dataset, the recognition accuracy of the proposed method reaches 88.2%, which verifies the effectiveness of the model.
-
Keywords:
- Low-light /
- Night video /
- Image enhancement /
- Attention mechanism /
- Behavior recognition
-
计量
- 文章访问数: 9
- HTML全文浏览量: 0
- PDF下载量: 5
下载:
