Multi-classifier integrated time sequence land utilization classification and change detection method

Abstract

The invention discloses a multi-classifier integrated time sequence land utilization classification and change detection method, which comprises the following steps of: obtaining a time sequence land utilization classification initial result and a land utilization change detection binary image of a target region based on a time sequence multispectral remote sensing image of the target region; performing spatial distribution and classification attribute consistency processing on the initial result of the post-period land utilization classification of the target area to obtain a post-period land utilization classification result of the target area, and repeating the processing to obtain a final time sequence land utilization classification result of the target area; and carrying out change type assignment on the land utilization change detection binary image to obtain land utilization classification change detection results of the target area in the previous and later periods, and carrying out processing again to obtain a final time sequence land utilization change detection result of the target area. According to the method, a land utilization classification result and a change detection result with consistent multi-period spatial distribution and attribute classification are finally realized.

Description

Technical Field

[0001] The invention belongs to the field of remote sensing technology, and in particular relates to a time series land use classification and change detection method integrated with multiple classifiers. Background Art

[0002] In recent years, with the increasing abundance of earth observation remote sensing satellite data resources, as well as the continuous development of cloud computing and machine learning technology, the application of remote sensing images in the extraction of land resource utilization and change information over a large area and long time series has been greatly promoted. The land use classification and change detection results obtained based on remote sensing images have also been widely used in various fields of national economic development, such as land and resources management and spatial planning, ecological and environmental protection, and disaster prevention and mitigation. Using multi-period multispectral remote sensing images to build a land resource utilization and change detection model can quickly extract land resource utilization and change information over a large spatial range, provide common information support for land resource utilization and change for business applications in various industries, and save manpower and material resources.

[0003] In terms of land use classification, remote sensing image monitoring technology can be used to extract long-term land surface water, vegetation, bare land and other coverage information and land resource utilization information such as buildings and roads. Traditional methods include supervised classification, unsupervised classification and object-oriented classification methods. In recent years, with the popularization of machine learning methods in remote sensing image classification applications, random forest (RF), support vector machines (SVM) and other methods have been widely used in remote sensing image land use classification, and have high classification accuracy.

[0004] In terms of land use change detection, long-term remote sensing images can be used to extract information on changes in land resource utilization in the target area, so as to grasp, evaluate and analyze the overall quality and change patterns of local land resource utilization, and provide support for rational land resource utilization planning and decision-making. Remote sensing monitoring methods for land resource utilization changes can be divided into two categories: post-classification change detection and pre-classification change detection. The advantage of the former is that it can obtain more accurate information on the type of change from "what to what", but there is a cumulative effect of time series remote sensing image classification errors, which will lead to large errors in post-classification change detection; the latter has a higher accuracy in detecting "change / non-change", but is not as accurate as the former in extracting the specific change type from "what to what", and there is confusion in the type of change.

[0005] At present, most of the land resource utilization and change detection methods are applied to land use classification or change detection separately: the method of extracting change information after land use classification will cause large change detection errors due to error accumulation, resulting in many pseudo changes caused by seasons, observation angles, lighting conditions, etc. being extracted; while the method of using direct change detection has a high "change / non-change" detection accuracy, but it is difficult to accurately distinguish the specific change type; if land use classification and land use change are extracted separately, there will often be inconsistencies between the land use change category, area, distribution and other information and the land use classification extraction information, which is not conducive to further statistics, analysis and evaluation of land resource utilization. In practical applications, it is usually necessary to consider and analyze land use classification data and change detection data in a coordinated manner in order to make more scientific and reasonable land resource utilization planning and decision-making. Therefore, it is necessary to develop a set of land use classification and change detection model methods to realize the integrated monitoring of land use classification and change, and effectively avoid the inconsistency of spatial distribution and classification attributes between land use classification and change information while having high detection accuracy. Summary of the invention

[0006] The technology of the present invention solves the problem: overcomes the shortcomings of the prior art, provides a time series land use classification and change detection method integrated with multiple classifiers, obtains a change detection result with higher accuracy by fusing the change detection results before and after classification and the land use classification results, and simultaneously obtains relatively accurate "from what to what" change type information; by processing the consistency of the final change detection results of the previous and subsequent periods and the land use classification results of the previous and subsequent periods, obtains a land use classification map and a land use change detection map with consistent time series spatial distribution and attribute classification, realizes the integrated processing of land use classification and change detection, and provides technical support for analyzing and evaluating the utilization of land resources and the law of change.

[0007] In order to solve the above technical problems, the present invention discloses a time series land use classification and change detection method based on multi-classifier integration, comprising:

[0008] Based on the time series multispectral remote sensing images of the target area, the land use classification model based on random forest classification, post-classification change detection, and change feature vector intensity threshold segmentation change detection are integrated to obtain the initial results of the time series land use classification of the target area and the binary map of land use change detection;

[0009] Based on the binary map of land use change detection and the land use classification results of the target area in the previous period, the initial results of the land use classification in the target area in the later period are processed for spatial distribution and classification attribute consistency to obtain the land use classification results of the target area in the later period; this process is repeated for the initial results of the time series land use classification of the target area to obtain the final time series land use classification results of the target area;

[0010] Based on the target area land use classification results after consistency processing before and after the period, the change type is assigned to the land use change detection binary map to obtain the target area land use classification change detection results before and after the period; this process is repeated on the target area time series land use change detection binary map to obtain the final target area time series land use change detection results.

[0011] In the above-mentioned multi-classifier integrated time series land use classification and change detection method, based on the time series multispectral remote sensing image of the target area, the land use classification model based on random forest classification, the change detection after classification, and the change feature vector intensity threshold segmentation change detection are integrated to obtain the initial results of the time series land use classification of the target area and the land use change detection binary map, including:

[0012] Based on the time series multispectral remote sensing images and DEM data of the target area, the feature data set and sample data set are constructed; on this basis, a land use classification model based on random forest classification is constructed to obtain the initial results of the time series land use classification of the target area;

[0013] Perform post-classification change detection, compare the initial land use classification results of the target area before and after the target area in the initial land use classification results of the target area time series, and obtain the time series change detection result a;

[0014] Based on the calculation of the spectral index of the time series multispectral remote sensing image of the target area, the change feature vector is constructed, the intensity of the change feature vector is calculated, and the time series change detection result b is obtained through adaptive threshold segmentation;

[0015] The time series change detection result a and the time series change detection result b are integrated to obtain the target area time series land use change detection binary map.

[0016] In the above-mentioned multi-classifier integrated time series land use classification and change detection method, feature data sets and sample data sets are constructed based on the time series multispectral remote sensing images and DEM data of the target area; and on this basis, a land use classification model based on random forest classification is constructed to obtain the initial results of the time series land use classification of the target area, including:

[0017] Based on the time series multispectral remote sensing images and DEM data of the target area, by calculating the surface reflectance, spectral index and terrain, a feature dataset including spectral surface reflectance, normalized vegetation index NDVI, soil adjusted vegetation index SAVI, normalized water index NDWI, texture feature principal components, DEM, slope, aspect and shadow landform images was constructed;

[0018] Based on the characteristic data set, with reference to higher-resolution remote sensing images of the same period, OpenStreetMap data, and combined with the surface cover characteristics of the target area, a land use classification system was established, and land use classification sample data was collected to construct a sample data set;

[0019] Based on the feature data set and sample data set, a land use classification model based on random forest classification was constructed to obtain the initial results of time series land use classification in the target area.

[0020] In the above-mentioned multi-classifier integrated time series land use classification and change detection method, post-classification change detection is performed, and the initial land use classification results of the target area before and after the target area are compared in the initial land use classification results of the target area time series to obtain the time series change detection result a, including:

[0021] The initial land use classification results of the target area time series are compared with the initial land use classification results of the target area before and after the target area, and the initial change detection results of the target area time series "from what to what" are obtained;

[0022] The obtained initial change detection result of the target area time series "from what to what" after classification is binarized to obtain the change detection result of the target area time series after binarization classification, that is, the time series change detection result a.

[0023] In the above-mentioned multi-classifier integrated time series land use classification and change detection method, based on the calculation of the spectral index of the time series multispectral remote sensing image of the target area, the change feature vector is constructed, the change feature vector intensity is calculated, and the time series change detection result b is obtained by adaptive threshold segmentation, including:

[0024] Combining the characteristics of the target area's land cover changes and the characteristics of the time series multispectral remote sensing images of the target area, a change feature vector including the spectral surface reflectance and spectral index is constructed;

[0025] The change feature vector intensity is calculated, and the change feature vector intensity is adaptively segmented by the OSTU threshold method to obtain the target area time series binary change detection result, that is, the time series change detection result b.

[0026] In the above-mentioned multi-classifier integrated time series land use classification and change detection method, the time series change detection result a and the time series change detection result b are fused to obtain a target area time series land use change detection binary map, including:

[0027] Perform an intersection operation on the time series change detection result a and the time series change detection result b to obtain the time series change detection result c;

[0028] The morphological transformation method is used to remove small areas and island holes from the time series change detection result c, and the target area time series binary change detection map, that is, the target area time series land use change detection binary map is obtained.

[0029] In the above-mentioned multi-classifier integrated time series land use classification and change detection method, based on the land use change detection binary map and the land use classification result of the target area in the previous period, the initial land use classification result of the target area in the later period is processed for spatial distribution and classification attribute consistency, and the land use classification result of the target area in the later period is obtained; this process is repeated for the initial land use classification result of the target area in the time series, and the final land use classification result of the target area in the time series is obtained, including:

[0030] The initial land use classification results of the target area in the initial period are post-processed to remove small spots and island holes, and a visual inspection of the post-classification processing is performed based on the multispectral remote sensing images of the target area in the initial period. The obvious misclassification and omission areas of the ground objects are manually edited, and the accuracy of the processed land use classification of the target area in the initial period is evaluated to obtain the land use classification of the target area in the initial period.

[0031] For the initial land use classification results of the target area in the non-starting period, based on the land use change detection binary map and the land use classification results of the target area in the previous period, the spatial distribution and classification attribute consistency processing is performed on the initial land use classification results of the target area in the later period, that is, the land use classification results of the later period are retained for the areas with changes in the later period, and the land use classification results of the previous period are used to assign values ​​to the areas without changes in the later period; this processing is repeated for the initial land use classification results of the target area time series to obtain the final land use classification results of the target area time series.

[0032] In the above-mentioned multi-classifier integrated time series land use classification and change detection method, based on the target area land use classification results after consistency processing in the previous and next periods, the land use change detection binary map is assigned a change type to obtain the target area land use classification change detection results before and after the previous and next periods; this process is repeated on the target area time series land use change detection binary map to obtain the final target area time series land use change detection results, including:

[0033] Based on the land use classification results of the target area after consistency processing in the previous and next periods, the change type is assigned to the land use change detection binary map, that is, the change type of "from what to what" is determined;

[0034] Repeat this process on the target area time series land use change detection binary map to obtain the final target area time series land use change detection result.

[0035] The present invention has the following advantages:

[0036] The present invention discloses a time series land use classification and change detection method integrated with multiple classifiers, which can realize the integrated monitoring of time series land use / cover classification and change by comprehensively applying remote sensing image feature extraction, machine learning, change vector analysis and adaptive threshold segmentation, and has the following advantages:

[0037] 1) The present invention utilizes multi-period multispectral remote sensing images to simultaneously obtain land use classification maps and land use change detection maps that are consistent with the spatial distribution and attribute classification of long-term series, and can realize the integrated detection of land use classification and change detection; by fusing the change detection results before and after classification, the change detection accuracy can be improved, and the change information of "from what to what" can be obtained at the same time; by processing the consistency of the change detection results of the previous and subsequent periods and the land use classification results of the previous and subsequent periods, the integrated detection of land use classification and change detection can be realized, which effectively reduces the error accumulation of change detection after classification, avoids the inconsistency problem between land use classification and change detection maps, and provides strong technical support for analyzing and evaluating the land resource utilization in the target area and understanding the law of land resource change.

[0038] 2) The present invention utilizes multi-period multispectral remote sensing images and digital elevation model (DEM) data to establish land use classification feature data sets, sample data sets, and change vector data sets, and constructs a land use classification and change detection model method, which can realize long-term land use classification and change detection; by using machine learning, remote sensing image post-classification detection, change vector analysis, adaptive threshold segmentation and other advanced technologies, the automation level of land use classification and change detection of time series remote sensing images is improved, and the constructed sample data sets can be reused, reducing the workload of manual interpretation, improving efficiency, and having versatility. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is an overall flow chart of a time series land use classification and change detection method of multi-classifier integration in an embodiment of the present invention. DETAILED DESCRIPTION

[0040] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments disclosed in the present invention will be further described in detail below with reference to the accompanying drawings.

[0041] Reference Figure 1 In this embodiment, the multi-classifier integrated time series land use classification and change detection method includes:

[0042] Step S1, based on the time series multispectral remote sensing image of the target area, integrates the land use classification model based on random forest classification, post-classification change detection, and change feature vector intensity threshold segmentation change detection to obtain the initial result of the time series land use classification of the target area and the land use change detection binary map.

[0043] In this embodiment, the specific implementation process of step S1 is as follows:

[0044] Sub-step S11, based on the time series multispectral remote sensing images and DEM data of the target area, construct a feature data set and a sample data set; and on this basis, construct a land use classification model based on random forest classification to obtain the initial results of the time series land use classification of the target area. Specifically: based on the time series multispectral remote sensing images and DEM data of the target area, by calculating the surface reflectance, spectral index, and terrain, construct a feature data set including spectral surface reflectance, normalized vegetation index NDVI, soil adjusted vegetation index SAVI, normalized water index NDWI, texture feature principal components, DEM, slope, aspect, and shadow landform images; based on the feature data set, refer to the higher resolution remote sensing images of the same period, OpenStreetMap data, combined with the surface cover characteristics of the target area, establish a land use classification system, and collect land use classification sample data to construct a sample data set; based on the feature data set and the sample data set, construct a land use classification model based on random forest classification to obtain the initial results of the time series land use classification of the target area.

[0045] Sub-step S12, perform post-classification change detection, compare the initial land use classification results of the target area before and after the target area in the initial land use classification results of the target area time series, and obtain the time series change detection result a. Specifically: use the differential comparison method to compare the initial land use classification results of the target area before and after the target area in the initial land use classification results of the target area time series, and obtain the initial post-classification change detection results of the target area time series "from what to what"; perform binarization processing on the obtained initial post-classification change detection results of the target area time series "from what to what", and obtain the binarized post-classification change detection results of the target area time series, that is, the time series change detection result a.

[0046] Sub-step S13, based on the spectral index calculation of the time series multispectral remote sensing image of the target area, constructs a change feature vector, calculates the intensity of the change feature vector, and obtains the time series change detection result b through adaptive threshold segmentation. Specifically: combining the characteristics of the change of the target area's ground cover and the spectral characteristics of the time series multispectral remote sensing image of the target area, constructs a change feature vector including the spectral surface reflectance and the spectral index; calculates the intensity of the change feature vector, and performs adaptive threshold segmentation operation on the intensity of the change feature vector through the OSTU threshold method to obtain the target area time series binary change detection result, that is, the time series change detection result b.

[0047] Sub-step S14, fusion of time series change detection result a and time series change detection result b, to obtain the target area time series land use change detection binary map. Specifically: the time series change detection result a and the time series change detection result b are subjected to intersection operation to obtain the time series change detection result c; the time series change detection result c is subjected to morphological transformation to remove small areas and island holes to obtain the target area time series binary change detection map, that is, the target area time series land use change detection binary map.

[0048] Step S2, based on the land use change detection binary map and the land use classification results of the target area in the previous period, the initial results of the land use classification in the target area in the later period are processed for spatial distribution and classification attribute consistency to obtain the land use classification results of the target area in the later period; this process is repeated for the initial results of the time series land use classification of the target area to obtain the final time series land use classification results of the target area.

[0049] In this embodiment, the specific implementation process of step S2 is as follows:

[0050] Sub-step S21, performs post-classification processing on the initial results of the land use classification of the target area in the initial period to remove small spots and island holes, performs visual inspection on the post-classification processing based on the multispectral remote sensing images of the target area in the initial period, manually edits the areas with obvious misclassification or omission of land objects, and conducts accuracy assessment on the processed land use classification of the target area in the initial period (the overall classification accuracy must reach more than 85%) to obtain the land use classification of the target area in the initial period.

[0051] Sub-step S22, for the initial land use classification results of the target area in the non-starting period, based on the land use change detection binary map and the land use classification results of the target area in the previous period, the spatial distribution and classification attribute consistency processing is performed on the initial land use classification results of the target area in the later period, that is, the land use classification results of the later period are retained for the areas that change in the later period, and the land use classification results of the previous period are used to assign values ​​to the areas that do not change in the later period; this processing is repeated for the initial land use classification results of the target area time series to obtain the final land use classification results of the target area time series.

[0052] S3, based on the target area land use classification results after consistency processing before and after the period, assign the change type to the land use change detection binary map to obtain the target area land use classification change detection results before and after the period; repeat this process on the target area time series land use change detection binary map to obtain the final target area time series land use change detection results.

[0053] In this embodiment, the specific implementation process of step S3 is as follows:

[0054] Sub-step S31, based on the target area land use classification results after consistency processing in the previous and subsequent periods, assign change types to the land use change detection binary map, that is, determine the change type of "from what to what".

[0055] Sub-step S32, repeating this process on the target area time series land use change detection binary map to obtain the final target area time series land use change detection result.

[0056] Based on the above steps S1 to S3, the land use classification results and change detection results with consistent spatial distribution and attribute classification in multiple periods are finally achieved. The benefits brought by this method include at least: reducing the error accumulation of change detection after classification, making it easier to determine the change type of change detection more accurately, effectively solving the inconsistency problem of time series land use classification and change detection in the statistical analysis of land resource utilization using time series remote sensing images, and having universality for land use classification and change detection of time series remote sensing images.

[0057] Based on the above embodiments, the time series land use classification and change detection method of the multi-classifier integration described in the present invention mainly includes the following parts:

[0058] (1) Initial classification of time series multispectral remote sensing images

[0059] In order to obtain the land use classification information and specific change type information of the target area, it is necessary to build a land use classification model for automatic land use classification in the target area. This method is based on time series multispectral remote sensing images and DEM data to construct a feature data set including spectral features, index features, texture features and topographic features. According to the surface cover and land use characteristics of the target area, a classification system and classification sample data set are established. On this basis, the random forest classification method is used to train the land use classification model to obtain the initial results of land use classification of time series multispectral remote sensing images in the target area.

[0060] (2) Post-processing of land use classification at the initial stage

[0061] Since the scheme of the present invention performs consistency processing of subsequent land use classification and change detection in previous and subsequent periods based on the land use classification results of the initial period, it is necessary to manually edit the land use classification results of the initial period to improve the classification accuracy. The initial results of the land use classification of the target area in the initial period are post-processed to remove small spots and island holes, and a visual inspection of the post-classification processing is performed based on the multispectral remote sensing image of the target area in the initial period, and the obvious misclassification and omission areas of the ground objects are manually edited, and the accuracy of the land use classification of the target area in the initial period after processing is evaluated (the overall classification accuracy must reach more than 85%), and the land use classification of the target area in the initial period is obtained.

[0062] (3) Change detection processing of time series multispectral remote sensing images

[0063] Limited by the imaging spectral range and spectral resolution of optical sensors, remote sensing images will show the phenomenon of "same object, different spectrum, same spectrum, different object". At the same time, due to the different atmospheric and lighting environments during imaging, the growth conditions of surface vegetation are different, and the land use classification of remote sensing images in multiple periods will produce classification errors, that is, confusion between different types of objects on the same image, and the same object on images of different periods is mistakenly classified as different types of objects. Therefore, the traditional land use change monitoring based on land use classification is easily affected by land use classification errors, resulting in cumulative errors. The intensity of the change feature vector can be used to determine whether the same pixel has changed in the previous and subsequent periods, and the direction of the change vector can be used to determine the change category of "from what to what". However, this method is more stringent in dealing with atmospheric correction, observation of solar altitude angle, soil moisture and seasonal differences in vegetation, and the identification of change types based on the direction of the change vector has the problem of using the same direction to measure changes between multiple types, thereby causing confusion in the change type. In response to the above problems, by integrating the results of the above two methods, a high-precision change detection result of "from what to what" is obtained. The specific steps are as follows:

[0064] 31) Change detection after land use classification

[0065] The differential comparison method is used for the initial land use classification results of the target area time series before and after the initial land use classification results to obtain the initial change detection results of the target area time series "from what to what" after classification; the obtained initial change detection results of the target area time series "from what to what" after classification are binarized to obtain the change detection results of the target area time series after binarization classification, that is, change detection result a.

[0066] 32) Change vector threshold segmentation change detection

[0067] In order to comprehensively utilize the spectral characteristics of the blue, green, red, near-infrared and other bands of the multispectral camera and the land use characteristics of the target area, appropriate change detection features are selected from the reflectivity, vegetation index, building index, soil index, texture and other features to construct a change feature vector. This embodiment selects the normalized vegetation index DNVI, the normalized water index DNWI, and the atmospheric impedance vegetation index ARVI as the construction of the change feature vector, and calculates the intensity of the change feature vector. The intensity of the change feature vector is adaptively segmented by the OSTU threshold method to obtain the target area time series binary change detection result, that is, the change detection result b. The calculation formulas involved include:

[0068]

[0069] ΔNDVI i =NDVI iT2 -NDVI iT1

[0070] ΔNDWI i =NDWI iT2 -NDWI iT1

[0071] ΔARVI i =ARVI iT2 -ARVI iT1

[0072]

[0073] Where ΔP i Represents the change feature vector of pixel i, through T 1 Phase and T 2 The difference calculation of the phase image feature vector is obtained; ΔNDVI i represents the normalized vegetation index of pixel i at T 1 Phase and T 2 Phase difference, ΔNDWI i represents the normalized water index of pixel i at T 1Phase and T 2 Phase difference, ΔARVI i The atmospheric impedance vegetation index of pixel i at T 1 Phase and T 2 Phase difference, NDVI iT2 Indicates that pixel i is at T 2 Normalized difference vegetation index value of the time phase, NDVI iT1 Indicates that pixel i is at T 1 Normalized difference vegetation index value of the phase, NDWI iT2 Indicates that pixel i is at T 2 Normalized difference water index value of the phase, NDWI iT1 Indicates that pixel i is at T 1 Normalized water index value of the phase, ARVI iT2 Indicates that pixel i is at T 2 The atmospheric impedance vegetation index value of the phase, ARVI iT1 Indicates that pixel i is at T 1 The atmospheric impedance vegetation index value of the phase, M i Indicates that pixel i is at T 1 Phase and T 2 Phase-varying eigenvector strength.

[0074] 33) Change detection result fusion

[0075] The change detection results a and b are obtained by using post-classification change detection and change vector threshold segmentation change detection, respectively. Both have a certain degree of pseudo change detection. To reduce pseudo change detection, this embodiment takes the intersection of the two change detection results, and then uses the morphological transformation method to remove small spots and small island holes from the fused change area to obtain a binary change detection map of the target area time series.

[0076] (4) Consistency processing of time series land use classification and change detection

[0077] In order to reduce the detection of pseudo-changes, the change detection results of this embodiment integrate the post-classification change detection and the change vector threshold segmentation change detection results. Therefore, there is a situation where the change detection area is inconsistent with the remote sensing image classification space, that is, there is a problem that the area with different land use classification categories in the previous and next periods in step (1) is inconsistent with the change area of ​​the change detection result. To solve this problem, it is necessary to perform spatial and attribute consistency processing on the land use classification results of the later period based on the land use change detection binary map, that is, for the initial land use classification results of the non-starting period of the target area, based on the land use change detection binary map and the land use classification results of the previous period of the target area, perform spatial distribution and classification attribute consistency processing on the initial land use classification results of the later period of the target area, that is, retain the land use classification results of the later period for the changed area of ​​the later period, and assign the land use classification results of the previous period to the non-changed area of ​​the later period; repeat this processing on the initial land use classification results of the target area time series to obtain the final land use classification results of the target area time series.

[0078] (5) Determination of change type of time series change detection graph

[0079] The binary map of land use change detection obtained through step (3) is that the pixel value of the changed area is 1, and the pixel value of the non-changed area is 0. In order to determine the change type of the changed area, it is necessary to assign the change type of the pixels in the changed area based on the land use classification map of the previous and next periods after consistency processing, that is, the land use category of the previous period is the category before the change of the changed pixel, and the land use category of the next period is the category after the change of the changed pixel, so as to determine the change from "what to what" and obtain the time series land use classification change detection result.

[0080] Based on the above-mentioned land use classification and change detection method model that integrates random forest classification and change vector threshold segmentation, remote sensing monitoring of land resource utilization and changes in a certain area was carried out. According to the time series remote sensing monitoring results, the local land resource utilization characteristics and change patterns were further analyzed, which provided support for the local land resource utilization planning and ecological environmental protection work, and also contributed to the maintenance of sustainable development.

[0081] Although the present invention has been disclosed as above in the form of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art may make possible changes and modifications to the technical solution of the present invention by using the methods and technical contents disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope of the technical solution of the present invention.

[0082] The contents not described in detail in the specification of the present invention belong to the common knowledge of the professionals in this field.

Claims

1. A time series land use classification and change detection method based on multi-classifier integration, characterized in that: include: Based on the time series multispectral remote sensing images of the target area, the land use classification model based on random forest classification, post-classification change detection, and change feature vector intensity threshold segmentation change detection are integrated to obtain the initial results of the time series land use classification of the target area and the binary map of land use change detection; Based on the binary map of land use change detection and the land use classification results of the target area in the previous period, the initial results of land use classification in the target area in the later period are processed for spatial distribution and classification attribute consistency, and the land use classification results of the target area in the later period are obtained; Repeat this process for the initial results of the time series land use classification of the target area to obtain the final results of the time series land use classification of the target area; Based on the land use classification results of the target area after consistency processing before and after the period, the change type is assigned to the land use change detection binary map to obtain the land use classification change detection results of the target area before and after the period; Repeat this process on the target area time series land use change detection binary map to obtain the final target area time series land use change detection result.

2. The method for time series land use classification and change detection based on multi-classifier integration according to claim 1 is characterized in that: Based on the time series multispectral remote sensing images of the target area, the land use classification model based on random forest classification, post-classification change detection, and change feature vector intensity threshold segmentation change detection are integrated to obtain the initial results of the time series land use classification of the target area and the land use change detection binary map, including: Based on the time series multispectral remote sensing images and DEM data of the target area, the feature data set and sample data set are constructed; on this basis, a land use classification model based on random forest classification is constructed to obtain the initial results of the time series land use classification of the target area; Perform post-classification change detection, compare the initial land use classification results of the target area before and after the target area in the initial land use classification results of the target area time series, and obtain the time series change detection result a; Based on the calculation of the spectral index of the time series multispectral remote sensing image of the target area, the change feature vector is constructed, the intensity of the change feature vector is calculated, and the time series change detection result b is obtained through adaptive threshold segmentation; The time series change detection result a and the time series change detection result b are integrated to obtain the target area time series land use change detection binary map.

3. The method for time series land use classification and change detection based on multi-classifier integration according to claim 2 is characterized in that: Construct feature datasets and sample datasets based on time series multispectral remote sensing images and DEM data of the target area; On this basis, a land use classification model based on random forest classification was constructed to obtain the initial results of time series land use classification in the target area, including: Based on the time series multispectral remote sensing images and DEM data of the target area, by calculating the surface reflectance, spectral index and terrain, a feature dataset including spectral surface reflectance, normalized vegetation index NDVI, soil adjusted vegetation index SAVI, normalized water index NDWI, texture feature principal components, DEM, slope, aspect and shadow landform images was constructed; Based on the characteristic data set, with reference to higher-resolution remote sensing images of the same period, OpenStreetMap data, and combined with the surface cover characteristics of the target area, a land use classification system was established, and land use classification sample data was collected to construct a sample data set; Based on the feature data set and sample data set, a land use classification model based on random forest classification was constructed to obtain the initial results of time series land use classification in the target area.

4. The method for time series land use classification and change detection based on multi-classifier integration according to claim 2 is characterized in that: Perform post-classification change detection, compare the initial land use classification results of the target area before and after the target area in the initial land use classification results of the target area time series, and obtain the time series change detection results a, including: The initial land use classification results of the target area time series are compared with the initial land use classification results of the target area before and after the target area, and the initial change detection results of the target area time series "from what to what" are obtained; The obtained initial change detection result of the target area time series "from what to what" after classification is binarized to obtain the change detection result of the target area time series after binarization classification, that is, the time series change detection result a.

5. The method for time series land use classification and change detection based on multi-classifier integration according to claim 2 is characterized in that: Based on the calculation of the spectral index of the time series multispectral remote sensing image of the target area, the change feature vector is constructed, the change feature vector intensity is calculated, and the time series change detection result b is obtained through adaptive threshold segmentation, including: Combining the characteristics of the target area's land cover changes and the characteristics of the time series multispectral remote sensing images of the target area, a change feature vector including the spectral surface reflectance and spectral index is constructed; The change feature vector intensity is calculated, and the change feature vector intensity is adaptively segmented by the OSTU threshold method to obtain the target area time series binary change detection result, that is, the time series change detection result b.

6. The method for time series land use classification and change detection based on multi-classifier integration according to claim 2 is characterized in that: The time series change detection result a and the time series change detection result b are integrated to obtain the target area time series land use change detection binary map, including: Perform an intersection operation on the time series change detection result a and the time series change detection result b to obtain the time series change detection result c; The morphological transformation method is used to remove small areas and island holes from the time series change detection result c, and the target area time series binary change detection map, that is, the target area time series land use change detection binary map is obtained.

7. The method for time series land use classification and change detection based on multi-classifier integration according to claim 1, characterized in that: Based on the binary map of land use change detection and the land use classification results of the target area in the previous period, the initial results of land use classification in the target area in the later period are processed for spatial distribution and classification attribute consistency, and the land use classification results of the target area in the later period are obtained; Repeat this process for the initial results of the time series land use classification of the target area to obtain the final results of the time series land use classification of the target area, including: The initial land use classification results of the target area in the initial period are post-processed to remove small spots and island holes, and a visual inspection of the post-classification processing is performed based on the multispectral remote sensing images of the target area in the initial period. The obvious misclassification and omission areas of the ground objects are manually edited, and the accuracy of the processed land use classification of the target area in the initial period is evaluated to obtain the land use classification of the target area in the initial period. For the initial land use classification results of the target area in the non-starting period, based on the land use change detection binary map and the land use classification results of the target area in the previous period, the spatial distribution and classification attribute consistency processing is performed on the initial land use classification results of the target area in the later period, that is, the land use classification results of the later period are retained for the areas with changes in the later period, and the land use classification results of the previous period are used to assign values ​​to the areas without changes in the later period; this processing is repeated for the initial land use classification results of the target area time series to obtain the final land use classification results of the target area time series.

8. The method for time series land use classification and change detection based on multi-classifier integration according to claim 1, characterized in that: Based on the land use classification results of the target area after consistency processing before and after the period, the change type is assigned to the land use change detection binary map to obtain the land use classification change detection results of the target area before and after the period; Repeat this process for the target area time series land use change detection binary map to obtain the final target area time series land use change detection results, including: Based on the land use classification results of the target area after consistency processing in the previous and next periods, the change type is assigned to the land use change detection binary map, that is, the change type of "from what to what" is determined; Repeat this process on the target area time series land use change detection binary map to obtain the final target area time series land use change detection result.

Images