CN103413123B - A kind of high-resolution remote sensing image airfield detection method based on conditional random field models - Google Patents

Abstract

The invention relates to an airport detection method for high-resolution remote sensing images based on a conditional random field model. The implementation process is as follows: first extracting dense SIFT features of high-resolution remote sensing images, and obtaining sparse coding of dense SIFT features on an over-complete dictionary, Then, a 4-connected graph is established within the 4-neighborhood range of each sparse code, and a conditional random field model is established on the 4-connected graph. According to the parameters of the conditional random field model learned by the Max‑margin algorithm, each sparse The probability value belonging to the airport target is encoded to obtain the airport target probability map, and finally the airport target probability map is thresholded to obtain the airport detection result. The invention performs airport detection in high-resolution remote sensing images, has the advantages of high accuracy and low false alarm rate, and has high application value.

Description

A Method of Airport Detection in High Resolution Remote Sensing Image Based on Conditional Random Field Model

technical field

The invention belongs to the technical field of remote sensing image processing, and in particular relates to a high-resolution remote sensing image airport detection method based on a conditional random field model.

Background technique

With the rapid development of satellite and sensor technology, high-resolution remote sensing images provide very rich information for target detection such as airports. Effective use of this information can improve the performance of airport detection. Most existing algorithms realize airport detection by using geometric information such as the linear features of airport runways, but pure geometric information cannot effectively distinguish airports from highways, rivers, artificial buildings, etc. Chinese patent application number 201110166001.X describes a "selective attention mechanism-based airport target detection and recognition method for remote sensing images", which uses SIFT features to characterize airports to achieve airport detection, but the background of high-resolution remote sensing images is complex. Only extracting SIFT local features for identification is easy to cause false detection. The Chinese Patent Application No. 201210282568.8 describes a "method for detecting airports in infrared remote sensing images based on sparse coding and visual saliency", which uses sparse coding guided by visual saliency to characterize airports. In low-resolution remote sensing images Good detection results have been obtained, but it is still not suitable for airport detection in high-resolution remote sensing images.

Contents of the invention

technical problem to be solved

In order to avoid the deficiencies of the prior art, the present invention proposes a high-resolution remote sensing image airport detection method based on a conditional random field model.

Technical solutions

A high-resolution remote sensing image airport detection method based on a conditional random field model, characterized in that it comprises the following steps:

Step 1: extract the dense SIFT feature x of the high-resolution remote sensing image, use the efficient sparse coding algorithm to learn the extracted dense SIFT feature to obtain an over-complete dictionary D, and obtain the sparse coding of the dense SIFT feature x on the over-complete dictionary D s(x,D);

Step 2: take each sparse code s(x, D) as a vertex, set up a 4-connected graph G within the 4 neighborhoods of the vertex, and set up a conditional random field model on the 4-connected graph G;

Step 3: Given the parameters of the conditional random field model, the LBP algorithm is used to deduce the probability value of each sparse code s(x, D) belonging to the airport, and obtain the airport target probability map;

Step 4: Use the adaptive threshold segmentation method to segment the airport target probability map to obtain the airport detection results.

The conditional random field model parameters in step 3 are learned by using the Max-margin algorithm.

Beneficial effect

The present invention proposes a high-resolution remote sensing image airport detection method based on a conditional random field model. First, the dense SIFT feature of the high-resolution remote sensing image is extracted, and the sparse coding of the dense SIFT feature on the over-complete dictionary is obtained, and then each A 4-connected graph is established within a sparsely coded 4-neighborhood, and a conditional random field model is established on the 4-connected graph. According to the parameters of the conditional random field model learned by the Max-margin algorithm, the LBP algorithm is used to deduce that each sparse code belongs to the airport. The probability value of the target can be obtained to obtain the airport target probability map. Finally, the airport target probability map can be segmented by threshold value to obtain the airport detection result.

Compared with the prior art, the present invention uses sparse coding to characterize the airport, which can effectively capture the structural characteristics of the airport, and put it in the conditional random field model to realize the detection of the airport. The present invention combines sparse coding and conditional random field The advantage of the model is that it can robustly represent the airport, and effectively utilize the spatial information of the surrounding neighborhood, which significantly reduces the false alarm rate of airport detection, and has a strong application value.

Description of drawings

Fig. 1 is the realization flowchart of the present invention.

Figure 2 is a high-resolution remote sensing image.

Figure 3 is an airport target probability map.

Figure 4 is a map of the airport detection results.

detailed description

Now in conjunction with embodiment, accompanying drawing, the present invention will be further described:

The hardware environment used for implementation is: Intel Xeon(R) CPU, E5504@2.0G 2.0G (2 processors) computer, 6.0GB memory, 1GB graphics card, and the running software environment is: Matlab R2012a, Windows7 64-bit operation system. We have realized the method that the present invention proposes with Matlab software. The high-resolution remote sensing images used in the experiment are all intercepted from Google Earth.

The present invention is specifically implemented as follows:

1. Learning through a complete and complete dictionary: intercept 20*3=60 high-resolution remote sensing images from Google Earth at three different viewing heights of 5 kilometers, 15 kilometers and 25 kilometers for 20 typical airports, and It is rotated and transformed every 45 degrees, and finally a total of 60*8=480 images are obtained, and these 480 images form a training set. Extract the dense SIFT feature of each training image, the image block size used for extraction is 64*64, and the sliding interval is 16. For the dense SIFT features of all the extracted training images, the efficient sparse coding algorithm is used to learn and obtain an over-complete dictionary D.

2. Extract the dense SIFT feature x in Figure 2, and obtain the sparse code s(x, D) of the dense SIFT feature x on the over-complete dictionary D;

3. Treat each sparse code s(x, D) as a vertex, and establish a 4-connected graph G within its 4-neighborhood range, and establish a conditional random field model on the 4-connected graph G;

4. Carry out airport labeling on each image in the training set to form the training set Ground Truth, and learn the conditional random field model parameters from the training set and its Ground Truth by the Max-margin algorithm;

5. According to the conditional random field model parameters obtained by learning, the probability value of each sparse code s(x, D) belonging to the airport is deduced by the LBP algorithm, so as to obtain the airport target probability map, that is, Figure 3;

6. Perform threshold segmentation on Figure 3 to obtain Figure 4 of the airport detection result.

Claims (2)

1. a high-resolution remote sensing image airfield detection method based on conditional random field models, it is characterised in that bag Include following steps:

Step 1: extract dense SIFT feature x of high-resolution remote sensing image, to the dense SIFT extracted Feature employing efficient sparse coding algorithm carries out study and obtained complete dictionary D, asks for dense SIFT feature x is at the sparse coding s (x, D) crossed on complete dictionary D；

Step 2: with each sparse coding s (x, D) as summit, set up 4 in 4 contiguous range on summit Connected graph G, set up the condition random field models on 4 connected graph G；

Step 3: specified criteria random field models parameter, is inferred each sparse coding s (x, D) by LBP algorithm Belong to the probit on airport, obtain airport target probability graph；

Step 4: use adaptive threshold fuzziness method to split airport target probability graph, obtains airport inspection Survey result.

A kind of high-resolution remote sensing image airport based on conditional random field models the most according to claim 1 is examined Survey method, it is characterised in that: the conditional random field models parameter of described step 3, by using Max-margin Algorithm Learning obtains.