A Batch Remote Sensing Image Preprocessing Method

Abstract

The invention discloses a batch remote sensing image preprocessing method. The method comprises the following steps: acquiring remote sensing images in batches and establishing a batch remote sensing image database, obtaining cloud-containing images by counting the spectral characteristics and texture characteristics of the remote sensing images; detecting cloud-containing images The range of the cloud domain and the cloud content of the cloud domain range are removed, and the cloud-containing image with the cloud content exceeding the cloud-containing threshold is removed; the cloud on the cloud-containing image is removed to obtain a cloud-free image; and the batch remote sensing image database is reconstructed. The invention is fast and efficient, has high precision, and has strong practical value and popularization significance; in the process of image preprocessing, the cloud content on the image is determined, and the cloud domain range on the image is automatically delineated and removed, and the The quality of the image is divided, and an accurate basis is given for image information extraction and data acquisition; the present invention adopts a decision tree method with high work efficiency, which has high efficiency and less human intervention, and can be used for batch selection of remote sensing images and remote sensing image information Extract research.

Description

technical field [0001] The invention belongs to the technical field of remote sensing and image processing, and in particular relates to a batch remote sensing image preprocessing method. Background technique [0002] In recent years, remote sensing images have been affected by clouds in the sky during the imaging process, which affects the interpretation and interpretation of ground objects. For remote sensing image information extraction, cloudy images will greatly reduce the utilization rate of remote sensing images, especially bring great inconvenience to remote sensing image preprocessing work, increase the difficulty of information extraction and processing, and reduce remote sensing image preprocessing. s efficiency. [0003] Since the 1990s, many domestic researchers in the field of remote sensing have carried out research in this area. A series of work on cloud detection has been carried out by threshold method, wavelet texture analysis method, object-oriented met...

AI Technical Summary

Problems solved by technology

[0004] For example, the threshold method was adopted in 2010 to realize the cloud-ground separation of FORMOSAT-2, VENDS, LANDSAT and SENTINEL-2 images according to the change of the reflectivity of clouds and ground objects with the wavelength, but it cannot guarantee that the threshold value is effective for the clouds on each image. Can achieve a good recognition effect

In 2007, non-sampling wavelet transform was used to remove thin clouds from panchromatic remote sensing images. The effect is better for single-scene remote sensing images, but for batch remote sensing data, the efficiency is not high and the adaptability is weak.

In 2011, the wavelet SCM method was used to extract texture features to identify clouds, and achieved certain results. It has good adaptability when clouds and snow are mixed, but it is affected by different types of sensors, and at the same time, due to the complexity of various ground objects on the image, the spatial distribution is not concentrated. , especially in the case of many types of image features, the effect of the cloud identified by this method is slightly worse

In 2012, the object-oriented Fmask algorithm was used to detect clouds. The accuracy of the algorithm is high, but it is difficult to distinguish between clouds and ice, especially in batch remote sensing image processing, and the efficiency is not high.

The support vector machine is used to detect and remove thick clouds in remote sensing images, and the method of statistical compensation is used to repair them. The accuracy is relatively high, but it is relatively weak in the detection of thin clouds.

In 2015, the BP neural network method was used for cloud detection. Because of its strong learning ability, understanding and identification of samples, the cloud detection effect is better, but the calculation amount is too large, which is not suitable for batch remote sensing image processing.

The above-mentioned work has achieved some achievements to varying degrees, but the working method is relatively single, and the effect is good and bad

The inaccurate recognition of cloud boundaries not only leads to a large number of invalid images being retrieved, but also causes huge waste

Images