【摘要】：針對(duì)低對(duì)比度、條紋噪聲、低空間分辨率等特點(diǎn)而導(dǎo)致的熱紅外圖像識(shí)別效果不佳問題,提出了一種港口目標(biāo)熱紅外遙感圖像特征提取與選擇方法,實(shí)現(xiàn)了一定情況下港口目標(biāo)的高精度分類。采用紋理、幾何等29個(gè)特征,通過評(píng)估器選擇最佳特征組合,并根據(jù)識(shí)別精度選擇最佳分類器,能生成熱紅外圖像港口目標(biāo)22個(gè)最佳分類特征,且具有一定的魯棒性。經(jīng)過參數(shù)優(yōu)化后的libSVM(一種支持向量機(jī))分類器分類精度較高;白天圖像比夜間圖像分類精度更高;像素值、灰度直方圖相關(guān)的一維和二維統(tǒng)計(jì)特征、局部二進(jìn)制模式特征、邊緣方向直方圖特征等與灰度和紋理相關(guān)的特征對(duì)港口目標(biāo)熱紅外圖像識(shí)別影響較大。
[Abstract]:In view of the poor recognition effect of thermal infrared image caused by low contrast, fringe noise and low spatial resolution, a method for feature extraction and selection of thermal infrared remote sensing image of port target is proposed. The high precision classification of port targets is realized under certain conditions. Using 29 features, such as texture and geometry, selecting the best combination of features through the evaluator and selecting the best classifier according to the recognition accuracy, the 22 best classification features of the thermal infrared image port can be generated and have certain robustness. After parameter optimization, libSVM (support Vector Machine) classifier has higher classification accuracy; daytime image classification accuracy is higher than night image classification accuracy; pixel value, gray histogram related one-dimensional and two-dimensional statistical feature, local binary pattern feature, pixel value, gray histogram correlation statistical feature, local binary pattern feature, The features of edge direction histogram such as grayscale and texture have great influence on the thermal infrared image recognition of port target.
【作者單位】： 信息工程大學(xué);國際關(guān)系學(xué)院;中國測繪科學(xué)研究院;
【分類號(hào)】：P237;TP751
，

本文編號(hào)：2291220