CN107657246B - A Building Detection Method Based on Multi-Scale Filtering Building Index in Remote Sensing Image - Google Patents

Abstract

The invention discloses a remote sensing image building detection method based on a multi-scale filtering building index. The invention is based on digital morphology and multi-scale filtering and realizes rapid and automatic detection of buildings in high-resolution remote sensing images. First, the brightness image is generated based on the maximum value of each band of the multi-band high-resolution remote sensing image; next, the brightness image is reconstructed based on the opening operation to generate an enhanced image; then, a series of median values of window sizes are used for the enhanced image Filter to obtain the multi-scale filtered differential image sequence; then average the sequence to obtain the multi-scale filtered building index and output the corresponding feature image. The size of the index can effectively represent the probability of buildings in remote sensing images, and based on this index, buildings in high-resolution remote sensing images can be effectively extracted.

Description

A Building Detection Method Based on Multi-Scale Filtering Building Index in Remote Sensing Image

technical field

The invention belongs to the technical field of image processing, and relates to a remote sensing image building detection method, in particular to a remote sensing image building detection method based on a multiscale filtering building index (Multiscale Filtering Building Index).

Background technique

Remote sensing is one of the important means of earth observation. Buildings are one of the important features in remote sensing images. The extraction of buildings from remote sensing images plays an important role in urban planning, census of geographical conditions, and GIS database update. However, in the face of massive high-resolution remote sensing data, intelligent interpretation technology has not improved to a satisfactory level. The automatic extraction of buildings from high-resolution remote sensing images mainly includes the following problems:

(1) Traditional methods based on spectral information are difficult to obtain good detection results;

Traditional remote sensing image information extraction is mostly based on spectral information, while high-resolution remote sensing images usually only have four bands: near-infrared, red, green, and blue. Compared with common low-resolution remote sensing images such as Landsat, the number of bands is significantly reduced. The spectral distinction information between objects and other ground objects is significantly reduced; in addition, high spatial resolution provides a large amount of detailed information of ground objects, the spectral difference within the same type of ground objects such as buildings increases, and the spectral difference between different ground objects decreases, resulting in The traditional spectrum-based remote sensing image information extraction methods are difficult to achieve good detection results in high-resolution remote sensing images. In fact, buildings can be seriously confused with features such as roads, bare land, and concrete.

(2) The complex and varied types of buildings have caused serious difficulties for high-precision identification;

Buildings are likely to be seriously confused with roads, bare land, and concrete ground in the limited band of high-resolution images; the structure and size of buildings vary greatly, and buildings in rural and suburban areas often occupy small areas and have irregular shapes , while urban buildings often occupy a large area and have different heights, and the architectural forms such as gymnasiums and commercial areas are too different from residential buildings. The above phenomena directly lead to the fact that although the supervised learning method has high precision, it often requires a lot of manpower and time to collect training samples of different types of buildings. The generalization ability of the classification model varies greatly, and the accuracy is greatly affected by the training set; Efficiency and adaptability vary greatly, and it is difficult to find a building extraction method that is suitable for rural, urban, and suburban areas at the same time, and the accuracy is often lower than supervised learning.

(3) In terms of data volume, algorithm efficiency and accuracy, it is still far from intelligent extraction;

At present, major geographic information production units such as surveying and mapping bureaus, urban planning bureaus, and geographic information centers still mainly use visual interpretation to manually outline buildings and other ground features, which consumes a lot of manpower and time and is inefficient. The main reason is that the existing intelligent remote sensing interpretation technology is difficult to meet the application requirements in terms of data volume, time, and accuracy. Since the 21st century, the launch of a large number of commercial high-resolution remote sensing satellites at home and abroad has made it no longer difficult to obtain TB-level high-resolution images every day. However, as summarized in (2), the current algorithms either take a long time, require manual collection of a large number of samples, or are weak in adaptability, making it difficult to effectively extract various types of buildings; current storage space and computer hardware are also difficult to do From synchronous and rapid extraction of buildings from massive remote sensing data, it is difficult for some existing high-performance equipment to be popularized in grassroots production units in the short term; in addition, existing research and references have shown that many building detection methods It performs well in the experimental area, but it is difficult to achieve high accuracy in other environments. Therefore, there is still a long way to go before automatic and intelligent extraction of ground objects in remote sensing images.

In summary, the automatic extraction of buildings from high-resolution remote sensing images is a difficult problem in target recognition, and it has rich application value, and it is still widely studied.

Contents of the invention

In order to solve the above technical problems, the present invention provides a remote sensing image building detection method based on multi-scale filtering building index. The multi-scale filtering building index is used to represent the probability that each pixel belongs to a building. The larger the index value, the greater the probability of belonging to a building.

The technical solution adopted in the present invention is: a method for detecting buildings in remote sensing images based on multi-scale filtering building index, characterized in that it includes the following steps:

Step 1: Construct a multi-scale filtering building index, and the specific implementation includes the following sub-steps;

Step 1.1: Generate brightness images for the acquired multi-spectral high-resolution remote sensing images;

Step 1.2: Perform enhancement processing on the brightness image to obtain an enhanced image;

Step 1.3: Obtain the differential image sequence of multi-scale median filtering;

Step 1.4: Generate multi-scale filtering building index;

Step 2: Automatically detect buildings in remote sensing images based on the multi-scale filtering building index, and the specific implementation includes the following sub-steps;

Step 2.1: Calculate the multi-scale filtering building index of each pixel for the obtained multi-spectral high-resolution remote sensing images;

Step 2.2: According to the multi-scale filtering building index corresponding to each pixel, set the threshold T, and the pixels greater than the threshold T are judged as buildings;

Step 2.3: Post-process the building image obtained in Step 2.2 to obtain the final detection result.

Compared with the prior art, the present invention has the following beneficial effects:

(1) An index representing buildings in high-resolution remote sensing images and a corresponding high-precision, fully automatic building detection method are provided.

(2) It can be effectively applied to the automatic and rapid extraction of buildings and the production of thematic information in the surveying, mapping, remote sensing and geographic information industries.

(3) It provides reference and reference for the rapid, high-precision and fully automatic intelligent extraction of thematic information of remote sensing images in the future.

Description of drawings

Fig. 1 is the flow chart of constructing multi-scale filtering building index according to the embodiment of the present invention;

Fig. 2 is a flow chart of automatic detection of remote sensing image buildings according to an embodiment of the present invention;

Fig. 3 is the common building detection method experimental result of the embodiment of the present invention and ground reality figure (sample area one),

Among them, (a) K-means, (b) gray level co-occurrence matrix, (c) MBI, d) MFBI, (e) ground truth map;

Fig. 4 is the common building detection method experimental result of the embodiment of the present invention and ground reality figure (sample area two),

Among them, (a) K-means, (b) gray level co-occurrence matrix, (c) MBI, d) MFBI, (e) ground truth map.

Detailed ways

In order to facilitate those of ordinary skill in the art to understand and implement the present invention, the present invention will be described in further detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the implementation examples described here are only used to illustrate and explain the present invention, and are not intended to limit this invention.

The present invention provides a remote sensing image building detection method based on multi-scale filtering building index, comprising the following steps:

Step 1: Construct a multi-scale filtering building index;

Please see Figure 1, the specific implementation includes the following sub-steps;

Step 1.1: Generate brightness images for the acquired multi-spectral high-resolution remote sensing images;

For the acquired multi-spectral high-resolution remote sensing images, the maximum value of each band of each pixel is obtained to generate a brightness image.

Step 1.2: Perform top-hat calculation based on reconstruction on the brightness image to obtain an enhanced image;

Step 1.3: Obtain the differential image sequence of multi-scale median filtering;

Perform median filtering on the enhanced image obtained in step 1.2 using a rectangular window with a side length from S _min to S _max and a distance of △S, respectively, and perform a difference between adjacent two of the obtained images to obtain a multi-scale median filter difference image sequence.

Step 1.4: Generate multi-scale filtering building index;

All the gray values corresponding to the pixels at the same position on the difference image sequence are averaged and normalized to 0-1 to obtain the multi-scale filter building index (MFBI). The larger the value of each pixel after this operation, the higher the probability of belonging to a building.

Step 2: Automatically detect buildings in remote sensing images based on the multi-scale filtering building index;

Please see Figure 2, the specific implementation includes the following sub-steps;

Step 2.1: Calculate the multi-scale filtering building index of each pixel for the obtained multi-spectral high-resolution remote sensing images;

In this embodiment, it is required to acquire as many bands as possible of the image. At least three bands of near-infrared, red, and green are required.

Step 2.2: According to the multi-scale filtering building index corresponding to each pixel, set the threshold T, and the pixels greater than the threshold T are judged as buildings;

Step 2.3: Post-processing the building image obtained in step 2.2 to obtain the final detection result;

For the building image obtained in step 2.2, in the spectral information part, false alarms caused by vegetation and water bodies are eliminated based on the NDVI and NDWI spectral information of each pixel; in the shape feature part, each connected region of the binary image is traversed, based on each The geometric features such as the area of the connected area and the aspect ratio are post-processed to eliminate false alarms and fill holes to obtain the final detection results.

The main innovation of the present invention is:

(1) Using the average value of the median difference sequence of each window to describe the probability that a pixel belongs to a building, a multi-scale filtering building index is proposed.

(2) Based on the multi-scale filtering building index, an automatic building detection method is studied.

In order to verify the effectiveness of this method, this example selects two representative remote sensing image sample areas with rich building types to carry out experiments. Using the K-means algorithm that only relies on spectral information for classification, the gray level co-occurrence matrix algorithm that only considers spatial structure information, and the morphological building index (MBI) that considers both spatial information and spectral information, a comparative experiment is carried out.

The detection results of K-means, gray-level co-occurrence matrix, morphological building index (MBI), multi-scale filtering building index (FMBI) and ground truth maps of sample area 1 and sample area 2 are shown in Fig. 3 and Fig. 4 respectively (a ) to (e) subfigures. It is worth noting that, based on the experimental results corresponding to the K-means method of classifying and extracting buildings based on spectral information, buildings are seriously confused with vegetation, roads, etc.; it is also difficult to distinguish buildings, Roads and some vegetation. Therefore, these two methods do not participate in the accuracy evaluation of building detection, and the quantitative accuracy evaluation and analysis are only carried out between this method and the Morphological Building Index (MBI). Quantitative accuracy evaluation uses three indicators: recall rate (Recall), precision rate (Precision) and overall accuracy (Overall Accuracy); quantitative analysis mainly considers these two methods (including post-processing) to complete the same sample area target extraction. time.

The above experiments were programmed and tested in the Visual Studio 2015 environment based on OpenCV 3.1.0.

Table 1 and Table 2 illustrate the detection accuracy of the morphological building index and the multi-scale filtering building index proposed in this paper; Table 3 illustrates the time required for the two methods to detect in two experimental areas.

Table 1 The detection accuracy of MBI and MFBI in sample area 1

Table 2 Detection accuracy of MBI and MFBI in sample area 2

Table 3 Calculation time of MBI and MFBI in sample area 1 and sample area 2 (unit: second)

MBI MFBI Area 1 21.8 1.97 Sample area two 23.5 2.49

From the innovations and experimental results of this method summarized above, it can be seen that the main advantages of this method are:

(1) This multi-scale filtering building index and the corresponding building extraction method fully combine the spectral information and spatial structure information of the image. In the process of calculating this index, the spectral information of the building is highlighted through the brightness image and the top-hat transformation of the open reconstruction, and the contrast with the surrounding objects is enlarged; the edge shape information of the building is highlighted through the difference sequence image ; The post-processing process of this method further takes into account the geometric shape characteristics such as area and aspect ratio. Therefore, a series of problems of traditional spectrum-based remote sensing image target extraction are fully overcome.

(2) The multi-scale filtering building index and the corresponding building extraction method have high detection accuracy for buildings of different shapes in the city. From the analysis in (1), it can be seen that this method fully considers the edge shape information of the building and the geometric shape characteristics such as area and aspect ratio. In addition, through a series of detection windows of different scales, this idea is different from the traditional template matching idea, and can effectively detect buildings of different shapes and sizes. This idea effectively solves the problem that the complex differences of building types affect the detection accuracy.

(3) The multi-scale filtering building index and the corresponding building extraction method have low algorithm cost and short calculation time. From the calculation time of the program in Table 3, it can be seen that for an area with a radius of about one kilometer, it only takes about 2 seconds to complete the building detection process based on this index. At the same time, the method has a high degree of automation, and only needs to manually input parameters without any interactive processing, thereby reducing manpower to the greatest extent. Therefore, this method can be effectively used for rapid and automatic extraction of buildings on large-area remote sensing images.

The above advantages can effectively solve the three difficulties in the automatic detection of buildings in high-resolution images proposed in the technical background section.

It should be understood that the parts not described in detail in this specification belong to the prior art.

It should be understood that the above-mentioned descriptions for the preferred embodiments are relatively detailed, and should not therefore be considered as limiting the scope of the patent protection of the present invention. Within the scope of protection, replacements or modifications can also be made, all of which fall within the protection scope of the present invention, and the scope of protection of the present invention should be based on the appended claims.

Claims (5)

1. A remote sensing image building detection method based on multi-scale filtering building indexes is characterized by comprising the following steps:

step 1: constructing a multi-scale filtering building index, and specifically realizing the multi-scale filtering building index comprises the following substeps;

Step 1.1: generating a brightness image aiming at the acquired multispectral high-resolution remote sensing image;

step 1.2: enhancing the brightness image to obtain an enhanced image;

step 1.3: acquiring a difference image sequence of multi-scale median filtering;

in step 1.3, median filtering is respectively performed on the enhanced image obtained in step 1.2 by using rectangular windows with side lengths from S _min to S _max and a distance of delta S, and differences are respectively performed on adjacent two of the obtained images to obtain a difference image sequence of multi-scale median filtering;

step 1.4: generating a multi-scale filtering building index;

step 2: the method comprises the following steps of automatically detecting a remote sensing image building based on a multi-scale filtering building index, and specifically realizing the method comprises the following substeps;

Step 2.1: calculating a multi-scale filtering building index of each pixel aiming at the acquired multi-spectral high-resolution remote sensing image;

step 2.2: setting a threshold T according to the multi-scale filtering building index corresponding to each pixel, and judging the pixel larger than the threshold T as a building;

Step 2.3: and (3) carrying out post-processing on the building image obtained in the step 2.2 to obtain a final detection result.

2. the method for detecting the buildings according to the remote sensing image based on the multi-scale filtering building index, which is characterized in that: in step 1.1, the maximum value of each wave band of each pixel is obtained, and a brightness image is generated.

3. The method for detecting the buildings according to the remote sensing image based on the multi-scale filtering building index, which is characterized in that: in step 1.2, the top hat operation based on reconstruction is carried out on the brightness image to obtain an enhanced image.

4. The method for detecting the buildings according to the remote sensing image based on the multi-scale filtering building index, which is characterized in that: in step 1.4, all gray values corresponding to pixels at the same positions on the difference image sequence are averaged and normalized to be between 0 and 1, and the multi-scale filtering building index is obtained.

5. The method for detecting the buildings according to the remote sensing image based on the multi-scale filtering building index, which is characterized in that: in step 2.3, in the spectral information part, the building image obtained in step 2.2 is subjected to false alarm removal caused by vegetation water body based on NDVI and NDWI spectral information of each pixel; and traversing the shape characteristic part to obtain each connected region of the binary image, performing post-processing based on the area and length-width ratio characteristics of each connected region, eliminating false alarms, filling holes and obtaining a final detection result.