论文部分内容阅读
针对低对比度、条纹噪声、低空间分辨率等特点而导致的热红外图像识别效果不佳问题,提出了一种港口目标热红外遥感图像特征提取与选择方法,实现了一定情况下港口目标的高精度分类。采用纹理、几何等29个特征,通过评估器选择最佳特征组合,并根据识别精度选择最佳分类器,能生成热红外图像港口目标22个最佳分类特征,且具有一定的鲁棒性。经过参数优化后的libSVM(一种支持向量机)分类器分类精度较高;白天图像比夜间图像分类精度更高;像素值、灰度直方图相关的一维和二维统计特征、局部二进制模式特征、边缘方向直方图特征等与灰度和纹理相关的特征对港口目标热红外图像识别影响较大。
Aiming at the poor recognition effect of thermal infrared images caused by low contrast, fringe noise and low spatial resolution, this paper proposes a method of feature extraction and selection of port-target thermal infrared remote sensing image, which realizes the high target of harbor under certain conditions Accuracy classification. Twenty-nine feature categories such as texture and geometry are selected, and the best feature set is selected by the evaluator. Based on the recognition accuracy, the best classifier is selected, which can generate 22 best classification features of thermal infrared image port and has certain robustness. The parameter-optimized libSVM (a support vector machine) classifier has higher classification accuracy; the daytime image is more accurate than the nighttime image classification; the pixel values, one-dimensional and two-dimensional statistics related to the gray histogram, the local binary mode feature , Edge-oriented histogram features and gray-scale and texture-related features have a great influence on the port thermal infrared image recognition.