Research On Classification And Recognition Algorithm Of High Resolution Remote Sensing Image On Ancient Villages

Due to the limitation of geography and the improper managements, ancient villages as an important carrier of Chinese traditional cultural heritage are dying out. With the enhancement of people's awareness of cultural heritage protection and the introduction of the relevant policies, protecting ancient villages is gradually attracting attention of scholars all over the world. Compared with traditional surveying and mapping based on artificial statistics, the high-efficiency of remote sensing technology and artificial intelligence provides a scientific and effective guarantee for the establishment of a platform on dynamic information management, protection and forecast of the ancient village.The classification and recognition of remote sensing image objects is the core content of the remote sensing analysis and interpretation. For high resolution remote sensing images with rich features and information, the traditional classification methods will result in a storm of issues, such as lower classification accuracy, spatial data redundancy and a great waste of resources. Furthermore, they can't meet the timeliness requirements in the practical application. Therefore, research on more efficient and more intelligent classification and recognition approaches of geographic information in high resolution remote sensing images is a current and future focus in the remote sensing information processing field.In this paper, taking the ancient village high resolution remote sensing images as the study object, according to “remote sensing image pre-processing-- remote sensing image segmentation-- ground objects classification and recognition” as the main line,the classification algorithms of ancient village high resolution remote sensing imagery were presented and explored.(1) For the shadows as an irresistible existence in the high resolution remote sensing images, combing with the characteristics of shadows in the ancient village high resolution remote sensing imagery, the shadows were processed with the pulse coupled neural networks close to the human visual properties. However, when the hue and brightness of some objects in the non-shadow regions are close to or even lower than that in shadow regions, employing traditional methods and PCNN always misreads the shadow regions whose intensity and hue falls in between that of the scene and objectives as non-shadow regions, moreover, entities with similar or darkerhue and similar or darker intensity may be regarded as shadows wrongly. In view of the above, the novel Double Thresholds Pulse Coupled Neural Networks Model with the mechanism of two different dynamic thresholds that continuously altered was proposed in the paper.(2) According to the diversity and complexity of the ground objects in the ancient village high resolution remote sensing imagery, a ground objects classification and recognition algorithm combing object-oriented thought with ensemble learning theory was developed. Firstly, multi-scale and multi-feature image segmentation was employed, and then the spectral features and the texture features as the input in the classification and recognition step were extracted. Ultimately, the final classification and recognition results were decided by the ensemble classifier which was constructed by multiple SVM member classifiers trained with the Ada Boost algorithm.