基于分区域特征提取的单张图像天气识别

Weather Classification of Single Image Based on Sub-region Feature Extraction

针对单张图像的传统天气识别方法准确率较低,以及对天空区域不明显的图像具有识别局限性的问题,提出基于分区域特征提取的天气识别方法;在含有晴天、多云、阴天、雾天4种天气的数据集中,对图像进行天际线分割,并识别天空区域与地面区域;分别提取图像中天空区域、地面区域和图像整体的纹理特征、形状特征及颜色特征,并利用随机森林分类模型进行训练和测试。结果表明,该方法对于单张图像天气识别的准确率为92.6%,可以准确地识别图像天气,具有较强的实用性和普适性。

Aiming at the problem that traditional weather recognition methods for single image had lower accuracies and recognition limitation for images without obvious sky region,a novel weather recognition method based on feature extraction in different regions was proposed.In a data set with four kinds of weather including clear sky,cloudy,overcast,and foggy days,the skyline of the image was segmented,and the sky area as well as the ground area were identified.Textural features,shape features and color features of the sky area,the ground area and the whole image were extracted respectively,which were trained and tested by using the random forest classification model.The results show that the accuracy of this method in single image weather recognition is 92.6%.It can recognize image weather accurately,and has stronger practicability and universality.