基于特征融合的极限学习机图像分类方法

Image Classification Method of ELM Combined with Feature Fusion

针对遥感图像分类结果受外界因素影响大、实时性较差等问题,提出一种融合图像卷积神经网络(Convolutional Neural Networks,CNN)特征和尺度不变特征变换(Scale Invariant Feature Transformation,SIFT)特征,结合极限学习机(Extreme Learning Machine,ELM)对遥感图像进行分类的方法。上述方法将CNN提取的整体特征和SIFT提取的局部特征相结合,并通过数据预处理降低了阴影、光照等外界因素对分类性能的影响;同时,通过图像信息熵改进的主成分分析(Principal Components Analysis,PCA)对融合后的特征降维,减少了数据维度的同时大大减少了数据降维过程中的计算量,提高了分类的实时性。最后,将得到的图像特征输入ELM分类器进行分类。用卫星遥感图像进行了仿真研究,结果表明该方法能有效提高图像分类准确率,具有良好的泛化性及实时性。

For the results of remote sensing image classification are largely affected by external factors and poor real-time performance,this paper proposes a method which fuses the CNN features and SIFT features of the images and combines ELM to classify the remote sensing images.This method was used to combine the local features extracted by CNN and the global features extracted by SIFT,and at the same time,to reduce the impact of external factors such as shadows and lighting on classification performance through data preprocessing.At the same time,the feature dimension reduction through PCA improved by image information entropy reduced the data dimension and the amount of calculation in the process of data dimensionality reduction,and improved the performance of classification real-time.Finally,the image features were input into the ELM classifier for classification.In this paper,satellite remote sensing images were used for simulation research.The results show that this method can effectively improve the accuracy of image classification,and has good generalization and real-time performance.