技术领域technical field
本发明涉及遥感图像目标检测识别领域，尤其是一种高效、高准确性的指定建筑区检测方法。在指定多类型不同时相建筑区的情况下，本发明方法比目前典型基于局部描述子匹配的指定建筑区检测方法具有更高计算效率及检测精度。The invention relates to the field of remote sensing image target detection and recognition, in particular to an efficient and high-accuracy detection method for designated building areas. In the case of specifying multiple types of building areas with different time phases, the method of the invention has higher calculation efficiency and detection accuracy than the current typical detection method for specifying building areas based on local descriptor matching.
背景技术Background technique
随着搭载平台及载荷能力的快速发展，遥感技术的应用领域逐步扩大。从遥感复杂大视场中高效地发现指定建筑物，在违章建筑监测、土地利用动态变化监测、反恐维稳、军事侦察等重要领域都有着迫切需求。With the rapid development of the carrying platform and load capacity, the application field of remote sensing technology is gradually expanding. Efficiently finding designated buildings from the complex and large field of view of remote sensing has an urgent need in important fields such as illegal building monitoring, land use dynamic change monitoring, anti-terrorism and stability maintenance, and military reconnaissance.
遥感图像具有视场覆盖范围广、场景复杂的特点，因此遥感图像中的指定建筑区检测方法需具有高时效性及高准确性。目前的算法大都只能实现对具有固定专有特征的某类建筑目标检测，如机场、油罐等。对于任意指定不具备专有特征的建筑区域不适用。此外，不同时相的遥感图像，在卫星平台拍摄角度、光照条件、天时、天候等因素的影响下，同一建筑区及其周围环境在遥感图像上的表征会有较大差异。为此，近年来学者们开始尝试采用局部描述子匹配技术对建筑区进行检测。一方面，由于在遥感大视场中地物类型丰富，此类方法在纹理丰富的非建筑区也会产生大量冗余的局部描述子，极大影响匹配检测效率及精度。另一方面，传统的局部描述子匹配方法并未针对建筑区特点进行有效设计，在复杂遥感大视场中的匹配效率及准确性难以得到保障。Remote sensing images have the characteristics of wide field of view coverage and complex scenes, so the detection method of designated building areas in remote sensing images needs to have high timeliness and high accuracy. Most of the current algorithms can only detect certain types of building targets with fixed proprietary features, such as airports and oil tanks. Not applicable to any designated building area that does not have an exclusive character. In addition, for remote sensing images of different time phases, under the influence of factors such as satellite platform shooting angle, lighting conditions, time of day, and weather, the representation of the same building area and its surrounding environment on remote sensing images will be quite different. For this reason, in recent years, scholars have begun to try to use local descriptor matching technology to detect built-up areas. On the one hand, due to the rich types of ground objects in the large field of view of remote sensing, such methods will also generate a large number of redundant local descriptors in non-building areas with rich textures, which greatly affects the efficiency and accuracy of matching detection. On the other hand, the traditional local descriptor matching method has not been effectively designed for the characteristics of built-up areas, and it is difficult to guarantee the matching efficiency and accuracy in complex remote sensing large fields of view.
发明内容Contents of the invention
本发明的目的在于针对以上方法在复杂大视场遥感图像指定建筑区检测时存在的局限性，提出一种层次化局部结构约束的遥感图像指定建筑区检测方法，充分利用建筑区域的局部结构信息，设计对建筑属性候选关键点及精确匹配点对的层次化约束策略，可大量减少冗余非建筑区的特征描述计算，并有效减少错误匹配点对，实现指定建筑区的快速可靠检测。The purpose of the present invention is to address the limitations of the above methods in the detection of designated building areas in complex and large-field remote sensing images, and propose a method for detecting designated building areas in remote sensing images with hierarchical local structure constraints, making full use of the local structure information of building areas , design a hierarchical constraint strategy for candidate key points of building attributes and exact matching point pairs, which can greatly reduce the feature description calculation of redundant non-building areas, and effectively reduce the wrong matching point pairs, so as to realize the rapid and reliable detection of designated building areas.
本发明针对以上叙述的情况，提出一种层次化局部结构约束的指定建筑区检测方法，目前国内尚未有此类方法的报道。Aiming at the situation described above, the present invention proposes a method for detecting designated building areas with hierarchical local structure constraints, but there is no report of such a method in China at present.
本发明提供的方法由初始关键点生成、结构模式描述的疑似建筑区关键点筛选、关键点局部结构相似性双层匹配及利用可靠匹配点对获取指定建筑区域四个主要部分组成。具体步骤如下：The method provided by the invention consists of four main parts: initial key point generation, key point screening of suspected building areas described by structural patterns, double-layer matching of key point local structural similarities, and use of reliable matching point pairs to obtain designated building areas. Specific steps are as follows:
第一步：初始关键点生成Step 1: Initial keypoint generation
本发明采用SIFT算法的关键点提取步骤，对指定建筑区切片和待检测整视场遥感图像进行初始关键点提取。The invention adopts the key point extraction step of the SIFT algorithm to extract the initial key points from the slice of the designated building area and the remote sensing image of the whole field of view to be detected.
第二步：结构模式描述的疑似建筑区关键点筛选Step 2: Screening of key points of suspected built-up areas described by structural patterns
为了减少待检测整场景中冗余的非建筑区关键点数量，提出基于结构模式描述的疑似建筑区关键点筛选方法。利用多级局部模式直方图特征，对上述步骤初始生成的关键点进行特征描述，并采用OC-SVM筛选出其中具有建筑属性的关键点。In order to reduce the number of redundant non-building key points in the whole scene to be detected, a method for screening key points of suspected building areas based on structural pattern description is proposed. Using the multi-level local pattern histogram feature, the key points initially generated in the above steps are characterized, and OC-SVM is used to filter out the key points with architectural attributes.
第三步：关键点局部结构相似性双层匹配Step 3: Two-layer matching of key point local structural similarity
基于上一步筛选获得疑似建筑区关键点后，首先基于特征欧氏距离进行快速粗匹配。由于表征同一建筑区的初匹配关键点在相同尺度上应包含相似的场景局部信息，特别是具有较高的局部骨架结构相似性。因此，本发明提出利用该特性进一步筛选出高结构相似性的匹配点对。After the key points of the suspected building area are obtained based on the previous step of screening, fast and rough matching is first performed based on the characteristic Euclidean distance. Since the initial matching key points representing the same building area should contain similar scene local information on the same scale, especially have high similarity of local skeleton structure. Therefore, the present invention proposes to use this feature to further screen out matching point pairs with high structural similarity.
第四步：利用可靠匹配点对获取指定建筑区域Step 4: Use reliable matching point pairs to obtain the designated building area
对上述经过相似性度量保留下来的可靠匹配点对，进行结构相似性排序，选取相似性最高的三对匹配对，利用仿射变换矩阵，将给定的参考图像的边界值带入变换模型，获取指定建筑区域。若上一步没有保留下可靠的匹配点对，则认为带测试图像中没有指定的建筑区域存在。Sequence the structural similarity of the reliable matching point pairs retained by the above similarity measurement, select the three matching pairs with the highest similarity, and use the affine transformation matrix to bring the boundary value of the given reference image into the transformation model, Get the specified building area. If no reliable matching point pairs are retained in the previous step, it is considered that there is no specified building area in the belt test image.
附图说明Description of drawings
图1是根据本发明的一个实施例的一种层次化局部结构约束的遥感图像指定建筑区检测方法的流程图。Fig. 1 is a flow chart of a method for detecting designated building areas in remote sensing images with hierarchical local structure constraints according to an embodiment of the present invention.
图2是匹配点邻域亮暗线骨架结构的提取的流程图。Fig. 2 is a flowchart of the extraction of bright and dark line skeleton structures in the neighborhood of matching points.
具体实施方式Detailed ways
以下说明如何具体实施本发明所提供方法。The following describes how to specifically implement the method provided by the present invention.
图1为是根据本发明的一个实施例的一种层次化局部结构约束的遥感图像指定建筑区检测方法的流程图，该方法包括：Fig. 1 is a flow chart of a method for detecting a designated building area in a remote sensing image with hierarchical local structure constraints according to an embodiment of the present invention, the method comprising:
第一步：初始关键点生成Step 1: Initial keypoint generation
本发明采用SIFT算法的关键点提取步骤，对指定建筑区切片和待检测整视场遥感图像进行初始关键点提取。其提取的步骤包括：极值点检测、关键点精确定位、关键点方向分配及局部描述子的生成。由于遥感视场地物复杂多样，整场景关键点提取会产生大量的待匹配关键点，包括建筑区、树木、绿地及裸地等区域。The invention adopts the key point extraction step of the SIFT algorithm to extract the initial key points from the slice of the designated building area and the remote sensing image of the whole field of view to be detected. The extraction steps include: extreme point detection, key point precise positioning, key point direction assignment and local descriptor generation. Due to the complexity and variety of objects in the remote sensing field of view, the key point extraction of the whole scene will generate a large number of key points to be matched, including areas such as construction areas, trees, green land, and bare land.
第二步：结构模式描述的疑似建筑区关键点筛选Step 2: Screening of key points of suspected built-up areas described by structural patterns
在待检测整视场遥感图像中，对上一步获取的SIFT关键点邻域进行多尺度局部模式直方图特征(MLPH)提取来描述局部模式特性；并采用OC-SVM区分出疑似建筑区与其他地物的关键点，过程如图1整流程中的结构模式描述的疑似建筑区关键点筛选部分所示。In the remote sensing image of the entire field of view to be detected, the multi-scale local pattern histogram feature (MLPH) is extracted from the SIFT key point neighborhood obtained in the previous step to describe the local pattern characteristics; and OC-SVM is used to distinguish suspected building areas from other The key points of the ground objects, the process is shown in the key point screening part of the suspected construction area described in the structural mode in the whole process of Figure 1.
第(2.1)步多尺度局部模式直方图特征描述Step (2.1) Multi-scale local pattern histogram feature description
基于上一步对待检测整视场遥感图像获取的初始SIFT关键点，提取每个关键点邻域的MLPH(多级局部模式直方图特征)来描述其局部模式特性。MLPH计算包括：Based on the initial SIFT key points obtained from the remote sensing image of the entire field of view to be detected in the previous step, the MLPH (multi-level local pattern histogram feature) of each key point neighborhood is extracted to describe its local pattern characteristics. MLPH calculations include:
图像的量化，包括对每个初始关键点，以其中心领域滑块，大小为h*h，若令中心像素为gc，产生模式矩阵：The quantization of the image includes, for each initial key point, the size of its center field slider is h*h. If the center pixel is gc , a pattern matrix is generated:
其中：gi为关键点邻域内的像素值，t为选定的阈值，Among them: gi is the pixel value in the neighborhood of the key point, t is the selected threshold,
矩阵分裂，包括将产生的模式矩阵分裂成三个子矩阵：正矩阵，等矩阵和负矩阵，三个矩阵分别定义为：Matrix splitting, including splitting the generated pattern matrix into three sub-matrices: positive matrix, equal matrix and negative matrix, the three matrices are defined as:
和模式直方图的产生，包括：and pattern histogram generation, including:
对于每个子矩阵，根据连通区域情况，生成对应的子直方图。其中为了减少直方图的维数并增加局部模式直方图的可识别性，需要对直方图进行合并处理，可通过公式：For each sub-matrix, the corresponding sub-histogram is generated according to the connected region. Among them, in order to reduce the dimension of the histogram and increase the recognizability of the local pattern histogram, the histogram needs to be merged, and the formula can be used:
vol(k)＝B×vol(k-1),k∈[2,L K]vol(k)=B×vol(k-1), k∈[2,L K]
其中，vol(k)表示已有直方图的每个区间的体积，B是控制区间增长速度的系数，一般取值为2，Among them, vol(k) represents the volume of each interval of the existing histogram, B is the coefficient for controlling the growth rate of the interval, and the general value is 2,
然后将三个直方图级联生成最后的局部模式直方图，其中，不同尺度对应不同的阈值t，Then the three histograms are concatenated to generate the final local pattern histogram, where different scales correspond to different thresholds t,
将不同尺度产生的单尺度局部模式直方图级联，得到最后的多尺度局部模式直方图，其中，阈值t的增长公式如下：The single-scale local pattern histograms generated at different scales are concatenated to obtain the final multi-scale local pattern histogram, where the growth formula of the threshold t is as follows:
tm＝T×tm-1,m∈[2,K,M]tm ＝T×tm-1 ,m∈[2,K,M]
s.t tM＜C＜tM+1st tM < C < tM+1
其中：T为控制阈值增长的系数，C为图像中的像素最大值，M为尺度的总数，最终每个关键点生成的MLPH特征共M×3×K维，不同地物关键点局部区域的MLPH特征具有明显差别，可用来区分疑似建筑区与其他地物的关键点。Among them: T is the coefficient that controls the growth of the threshold, C is the maximum value of pixels in the image, M is the total number of scales, and finally the MLPH features generated by each key point have a total of M×3×K dimensions, and the local area of different key points The MLPH features have obvious differences and can be used to distinguish the key points of suspected building areas and other ground objects.
第(2.2)步基于OC-SVM的建筑区关键点辨识Step (2.2) Key point identification of building area based on OC-SVM
基于步骤(2.1)提取的关键点邻域的MLPH特征，利用具有良好单类辨识特性的OC-SVM分类器构建建筑属性关键点邻域的样本概率分布二值模型，其中筛选阶段利用训练好的OC-SVM模型，判断待测关键点邻域样本属于建筑类或非建筑类。Based on the MLPH features of the key point neighborhood extracted in step (2.1), the OC-SVM classifier with good single-class identification characteristics is used to construct the sample probability distribution binary model of the key point neighborhood of building attributes, in which the trained The OC-SVM model judges whether the neighborhood sample of the key point to be tested belongs to the building category or the non-building category.
第三步：关键点局部结构相似性双层匹配Step 3: Two-layer matching of key point local structural similarity
针对上一步对待检测整视场筛选得到的具有建筑属性的关键点，先利用关键点的SIFT特征欧氏距离获取初始匹配点对。再通过计算初始匹配点对的局部骨架结构相似性筛选出高可靠性的精匹配点对。For the key points with architectural attributes screened in the entire field of view to be detected in the previous step, first use the SIFT feature Euclidean distance of the key points to obtain the initial matching point pairs. Then, the fine matching point pairs with high reliability are screened out by calculating the similarity of the local skeleton structure of the initial matching point pairs.
第(3.1)步关键点粗匹配Step (3.1) key point rough matching
对参考指定建筑区域图像提取SIFT关键点，对大视场待检测遥感图像按上述步骤提取筛选后的建筑属性关键点，并计算筛选后关键点对应周围邻域的SIFT匹配特征。然后基于特征的欧式距离进行粗匹配，获取匹配点对：Extract SIFT key points from the image of the reference designated building area, and extract the key points of the filtered building attributes from the remote sensing image with a large field of view to be detected according to the above steps, and calculate the SIFT matching features of the filtered key points corresponding to the surrounding neighborhood. Then perform rough matching based on the Euclidean distance of the feature to obtain matching point pairs:
其中，Ri和Sj分别为参考图像和遥感测试图像中的SIFT特征描述子，d(Ri,Sj)是与Ri的所有欧式距离中的最小值，d(Ri,Sk)为次小值，thr为设定的阈值，一般取值为0.8。Among them, Ri and Sj are the SIFT feature descriptors in the reference image and the remote sensing test image respectively, d(Ri ,Sj ) is the minimum value of all Euclidean distances to Ri , d(Ri ,Sk ) is the next smallest value, and thr is the set threshold, which is generally 0.8.
第(3.2)步基于局部骨架结构相似性的关键点精匹配Step (3.2) Key point fine matching based on local skeleton structure similarity
把参考图像中经过粗匹配后的匹配关键点集合定义为P＝{pi},i＝1,L N，其中：N为粗匹配后的匹配对数，其中每个关键点包含位置、尺度和方向三项信息。匹配点邻域亮暗线骨架结构提取的流程如图2所示。The set of matching key points after rough matching in the reference image is defined as P={pi }, i=1, LN, where: N is the number of matching pairs after rough matching, and each key point includes position, scale and Orientation three items of information. The process of extracting the skeleton structure of bright and dark lines in the neighborhood of matching points is shown in Figure 2.
对上述粗匹配后的关键点，在其对应尺度和位置上计算关键点周围邻域的亮暗线骨架结构。其中，邻域半径选用对应描述子的半径，由下式确定：For the above key points after rough matching, calculate the bright and dark line skeleton structure of the neighborhood around the key point at its corresponding scale and position. Among them, the radius of the neighborhood is selected from the radius of the corresponding descriptor, which is determined by the following formula:
其中：d＝4，σoct为关键点的组内尺度，Among them: d=4, σoct is the intra-group scale of key points,
对关键点邻域提取亮暗线骨架，并计算其骨架结构来表征该局域的结构特性。包括：The bright and dark line skeleton is extracted for the key point neighborhood, and its skeleton structure is calculated to characterize the structural characteristics of the local area. include:
先计算邻域的平滑自适应二值分割图像BM:First calculate the smooth adaptive binary segmentation image BM of the neighborhood:
BM＝(Otsu(MoP)ob)·bBM=(Otsu(MoP)ob)·b
对经过Otsu自适应阈值分割得到的二值图像进行开闭运算，去除二值图像的点噪声并填充孔洞，得到体现主要轮廓的的二值图像数据，其中：定义fob为用结构元素b对图像f进行形态学开运算，f·b用结构元素b对图像f进行形态学闭运算。Open and close the binary image obtained by Otsu adaptive threshold segmentation, remove the point noise of the binary image and fill the hole, and obtain the binary image data that reflects the main outline, where: define fob as using the structural element b to image f performs the morphological opening operation, and f·b performs the morphological closing operation on the image f with the structural element b.
对BM关键点邻域前景部分提取亮线骨架，并计算其骨架结构density_br：Extract the bright line skeleton for the foreground part of the BM key point neighborhood, and calculate its skeleton structure density_br:
其中，定义Mor(BW,opt)为对二值图像进行形态学骨骼操作，当opt＝skel时，提取骨骼，当opt＝spur时，去除骨骼中的毛刺，其中row(·)为图像的行数，col(·)为列数，相应地，暗线骨架的提取区域为对应BM的背景部分，计算其骨架结构density_dr：Among them, Mor(BW,opt) is defined as the morphological bone operation on the binary image. When opt=skel, the bone is extracted. When opt=spur, the burr in the bone is removed, where row(·) is the row of the image number, col(·) is the number of columns, correspondingly, the extraction area of the dark line skeleton is the background part of the corresponding BM, and its skeleton structure density_dr is calculated:
从而，对参考图像和测试图像中的每个匹配点，同时提取了亮线骨架结构density_br和暗线骨架结构density_dr。Thus, for each matching point in the reference image and the test image, the bright line skeleton structure density_br and the dark line skeleton structure density_dr are simultaneously extracted.
(1)计算匹配点对的骨架结构相似性：(1) Calculate the skeleton structure similarity of matching point pairs:
针对粗匹配后参考图像与测试图像中的匹配点具有“多对一”的特点，本发明以测试图中匹配关键点为基准，遍历计算每个匹配点对的亮暗线骨架结构相似性，其中把参考图像中的关键点pri与测试图像中对应的一个匹配点psi1的亮暗线骨架结构相似性度量定义为:In view of the "many-to-one" feature of the matching points in the reference image and the test image after rough matching, the present invention uses the matching key points in the test image as a benchmark to traverse and calculate the similarity of the bright and dark line skeleton structure of each pair of matching points, where The similarity measure of the bright and dark line skeleton structure between the key point pri in the reference image and a corresponding matching point psi1 in the test image is defined as:
其中测试图中关键点邻域区域与参考图中的骨架结构越相似该结构Among them, the more similar the key point neighborhood area in the test image is to the skeleton structure in the reference image, the structure
值差别越小。The value difference is smaller.
(3)利用相似性筛选可靠匹配点对：(3) Use similarity to screen reliable matching point pairs:
由于需要剔除的不可靠匹配点对包括“多对一”及不稳定的匹配对，因此本发明提出利用骨架结构相似性对可靠匹配对进行筛选，当在“多对一”的参考图像中的一个关键点pri对应测试图像中的多个匹配点psi1,L,且N0为匹配点的个数且N0∈[1,N]时，分别得到对应的N0个骨架结构相似性S(pri,psi1),L,并求其中的最大值S(pri,psir)＝max(S(pri,psi1),L,其对应N0个数据中的第r个。此时认定关键点psir是正确且唯一的与关键点pri对应的匹配点且骨架结构相似性S的大小表征了匹配的两个关键点邻域的结构相似性，并设定阈值T，当第o对匹配对满足：Since the unreliable matching point pairs that need to be eliminated include "many-to-one" and unstable matching pairs, the present invention proposes to use the skeleton structure similarity to screen reliable matching pairs. When the "many-to-one" reference image A key point pri corresponds to multiple matching points psi1 ,L, And N0 is the number of matching points and when N0 ∈ [1,N], the corresponding N0 skeleton structure similarities S(pri ,psi1 ),L, And find the maximum value S(pri ,psir )=max(S(pri ,psi1 ),L, It corresponds to the r-th among the N0 data. At this time, it is determined that the key point psir is the correct and unique matching point corresponding to the key pointpri and the size of the skeleton structure similarity S represents the structural similarity of the two key point neighborhoods matched, and the threshold T is set , when the oth matching pair satisfies:
S(pro,pso)＜TS(pro ,pso )＜T
认为该匹配对是不稳定匹配点对，进行剔除，最后获得可靠的匹配对，进行后续关键点的精匹配。如果得到的全部S值都小于阈值T，则认定测试图中没有待检测的建筑区域。It is considered that the matching pair is an unstable matching point pair, and it is eliminated, and finally a reliable matching pair is obtained, and the subsequent fine matching of key points is carried out. If all the obtained S values are smaller than the threshold T, it is determined that there is no building area to be detected in the test map.
第四步：利用可靠匹配点对获取指定建筑区域Step 4: Use reliable matching point pairs to obtain the designated building area
由步骤(3.2)中获得的匹配对中选取可靠性最高，即骨架结构相似性S最大的三个匹配对作为仿射变换获取指定建筑区域，包括对参考图像中的三个匹配点表示为：(x1,y1)，(x2,y2)，(x3,y3)和测试图像中对应的匹配点表示为：(x1′，y1′)，(x2′，y2′)，(x3′，y3′)，代入仿射变换的模型：From the matching pairs obtained in step (3.2), select the three matching pairs with the highest reliability, that is, the three matching pairs with the largest skeleton structure similarity S as affine transformation to obtain the designated building area, including the three matching points in the reference image expressed as: (x1 , y1 ), (x2 , y2 ), (x3 , y3 ) and the corresponding matching points in the test image are expressed as: (x1 ′, y1 ′), (x2 ′, y2 ′), (x3 ′, y3 ′), substituted into the model of affine transformation:
利用该仿射变换矩阵对给定的参考图像的边界值带入变换模型，得到指定建筑区域在测试图像中的位置的边界值，获取到指定的建筑区域。Using the affine transformation matrix to bring the boundary value of the given reference image into the transformation model, obtain the boundary value of the position of the designated building area in the test image, and obtain the designated building area.
本发明与现有检测方法相比具有以下优点:Compared with existing detection methods, the present invention has the following advantages:
(1)针对实际复杂大视场应用场景下，已有检测方法普遍存在效率低、匹配检测准确性差的问题。本发明针对遥感图像指定建筑区检测，提出一种层次化局部结构约束的检测架构，该方法充分利用建筑区域的局部结构信息，设计对建筑属性候选关键点及精确匹配点对的层次化约束策略。本发明检测方法可大量减少冗余非建筑区的特征描述计算，并有效减少错误匹配点对，实现指定建筑区的可靠检测。(1) For the actual complex and large field of view application scenarios, the existing detection methods generally have the problems of low efficiency and poor matching detection accuracy. Aiming at the detection of designated building areas in remote sensing images, the present invention proposes a detection framework of hierarchical local structural constraints. The method makes full use of the local structural information of building areas to design a hierarchical constraint strategy for candidate key points of building attributes and accurate matching point pairs. . The detection method of the invention can greatly reduce the feature description calculation of the redundant non-building area, effectively reduce the wrong matching point pairs, and realize the reliable detection of the designated building area.
(2)在关键点生成阶段，本发明提出基于结构模式描述的疑似建筑区关键点筛选方法。利用多级局部模式直方图特征(MLPH)，对初始关键点进行局部模式特征描述，然后采用OC-SVM筛选出具有建筑属性的关键点。该方法可以从初始图像数量巨大的初始关键点中，有效筛选出任务相关的建筑属性关键点，极大减少了后续对非建筑区计算描述特征和匹配的运算量，提高了算法效率。此外，通过OC-SVM仅关注于建筑区关键点的特性，克服了非建筑区关键点产生于各类地物，特征属性丰富多样难点，提高了建筑属性关键点辨识的准确性，从而提高了整个匹配检测算法的准确性。(2) In the key point generation stage, the present invention proposes a key point screening method for suspected building areas based on structural pattern description. Using multi-level local pattern histogram feature (MLPH), the initial keypoints are described with local pattern features, and then OC-SVM is used to filter out keypoints with architectural attributes. This method can effectively filter out task-related building attribute key points from the initial key points with a large number of initial images, which greatly reduces the amount of calculations for subsequent calculation of description features and matching for non-building areas, and improves the efficiency of the algorithm. In addition, through OC-SVM only focusing on the characteristics of key points in the building area, it overcomes the difficulty that key points in non-building areas are generated from various types of ground objects, and the feature attributes are rich and diverse, which improves the accuracy of key point identification of building attributes, thereby improving The accuracy of the entire match detection algorithm.
(3)在关键点匹配阶段，本发明提出基于关键点局部结构相似性的双层匹配方法，第一层先利用关键点的SIFT特征欧氏距离获取初始匹配点对。第二层计算粗匹配后关键点周围邻域的暗亮线骨架结构来表征其局部结构，并对具有较高局部结构相似性的关键点进行精匹配。该方法针对建筑区关键点的局域特性，通过粗精两层的层次化匹配设计，达到筛选出可靠匹配点对的目的，提高了匹配检测算法的准确性。(3) In the key point matching stage, the present invention proposes a two-layer matching method based on the local structural similarity of the key points. The first layer first uses the SIFT feature Euclidean distance of the key points to obtain the initial matching point pair. The second layer calculates the skeleton structure of dark and bright lines in the neighborhood around the keypoint after rough matching to characterize its local structure, and fine-matches keypoints with higher local structure similarity. According to the local characteristics of the key points in the building area, this method achieves the purpose of screening out reliable matching point pairs through the hierarchical matching design of coarse and fine layers, and improves the accuracy of the matching detection algorithm.
(4)本发明提出一种基于局部骨架结构的建筑区关键点相似性度量方法。在目前经典的关键点相似性度量方法中，没有针对建筑区关键点的特性进行针对性的相似性度量方法设计。本发明先采用亮暗线骨架结构来描述匹配关键点的邻域结构特征，再基于该特征计算匹配点对的骨架结构相似性。该相似性度量方法通过亮暗线骨架对局部建筑区的主要结构实现了有效刻画，并通过计算亮暗线骨架的密度加权平均，实现了建筑区关键点相似性的有效度量。(4) The present invention proposes a method for measuring the similarity of key points in building areas based on local skeleton structures. In the current classical key point similarity measurement methods, there is no targeted similarity measurement method design for the characteristics of the key points in the built-up area. The present invention first uses bright and dark line skeleton structure to describe the neighborhood structure feature of matching key points, and then calculates the skeleton structure similarity of matching point pairs based on the feature. The similarity measurement method achieves an effective description of the main structure of the local building area through the bright and dark line skeleton, and realizes the effective measurement of the similarity of the key points of the building area by calculating the density-weighted average of the bright and dark line skeleton.
(5)目前的算法大都只能实现对具有固定专有特征的某类建筑目标检测，对于任意指定不具备专有特征的建筑区域不适用。本发明可针对任意指定的建筑区域，通过初始关键点生成、结构模式描述的疑似建筑区关键点筛选、关键点局部结构相似性双层匹配、及利用可靠匹配点对获取指定建筑区域四个主要步骤，实现任意指定建筑区的可靠匹配检测，并具有对不同时相的遥感图像，在卫星平台拍摄角度、光照条件、天时、天候等因素的影响下的良好适应性。(5) Most of the current algorithms can only realize the detection of certain types of building objects with fixed and exclusive features, and are not suitable for arbitrarily designated building areas without exclusive features. The present invention can aim at any designated building area, through the generation of initial key points, the screening of key points in suspected building areas described by structural patterns, the double-layer matching of key point local structure similarities, and the use of reliable matching point pairs to obtain four main points of designated building areas. Steps to achieve reliable matching detection of any designated building area, and have good adaptability to remote sensing images of different time phases, under the influence of satellite platform shooting angle, lighting conditions, weather, weather and other factors.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710854158.9ACN107622239B (en) | 2017-09-15 | 2017-09-15 | A Method for Detection of Specified Building Areas in Remote Sensing Images Constrained by Hierarchical Local Structure |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710854158.9ACN107622239B (en) | 2017-09-15 | 2017-09-15 | A Method for Detection of Specified Building Areas in Remote Sensing Images Constrained by Hierarchical Local Structure |
| Publication Number | Publication Date |
|---|---|
| CN107622239A CN107622239A (en) | 2018-01-23 |
| CN107622239Btrue CN107622239B (en) | 2019-11-26 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710854158.9AActiveCN107622239B (en) | 2017-09-15 | 2017-09-15 | A Method for Detection of Specified Building Areas in Remote Sensing Images Constrained by Hierarchical Local Structure |
| Country | Link |
|---|---|
| CN (1) | CN107622239B (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108572285B (en)* | 2018-03-26 | 2019-12-31 | 北京航空航天大学 | A high-speed optocoupler screening method based on low-frequency broadband noise |
| CN112182125B (en)* | 2020-09-14 | 2022-07-05 | 中国科学院重庆绿色智能技术研究院 | Business gathering area boundary identification system |
| CN114612696B (en)* | 2020-12-04 | 2024-12-03 | 北京达佳互联信息技术有限公司 | Target image screening method, device, electronic device and storage medium |
| CN112614139B (en)* | 2020-12-17 | 2022-09-16 | 武汉工程大学 | Conveyor belt ore briquette screening method based on depth map |
| CN112801206B (en)* | 2021-02-23 | 2022-10-14 | 中国科学院自动化研究所 | Image key point matching method based on depth map embedded network and structure self-learning |
| CN113378642B (en)* | 2021-05-12 | 2022-05-06 | 三峡大学 | A method for detecting illegally occupied buildings in rural areas |
| CN114266703A (en)* | 2022-03-03 | 2022-04-01 | 凯新创达（深圳）科技发展有限公司 | Image splicing method and system |
| CN114926812B (en)* | 2022-05-09 | 2025-06-27 | 中国科学院上海微系统与信息技术研究所 | An adaptive focusing positioning target detection method |
| CN119478712B (en)* | 2025-01-16 | 2025-04-01 | 中国建筑西南设计研究院有限公司 | Ultra-large-span high-precision special-shaped structure construction monitoring method, device and medium |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101894261A (en)* | 2010-06-29 | 2010-11-24 | 武汉大学 | A Multi-contrast Mode Histogram Texture Descriptor Extraction Method |
| CN103761523A (en)* | 2014-01-02 | 2014-04-30 | 北京理工大学 | Automatic identification and tracking method for airborne remote sensing video in specific man-made area |
| CN104077775A (en)* | 2014-06-28 | 2014-10-01 | 中国科学院光电技术研究所 | Shape matching method and device combining skeleton feature points and shape context |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3345129A4 (en)* | 2015-08-31 | 2019-07-24 | Cape Analytics, Inc. | SYSTEMS AND METHODS FOR ANALYZING REMOTE DETECTION IMAGING |
| US10055839B2 (en)* | 2016-03-04 | 2018-08-21 | Siemens Aktiengesellschaft | Leveraging on local and global textures of brain tissues for robust automatic brain tumor detection |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101894261A (en)* | 2010-06-29 | 2010-11-24 | 武汉大学 | A Multi-contrast Mode Histogram Texture Descriptor Extraction Method |
| CN103761523A (en)* | 2014-01-02 | 2014-04-30 | 北京理工大学 | Automatic identification and tracking method for airborne remote sensing video in specific man-made area |
| CN104077775A (en)* | 2014-06-28 | 2014-10-01 | 中国科学院光电技术研究所 | Shape matching method and device combining skeleton feature points and shape context |
| Publication number | Publication date |
|---|---|
| CN107622239A (en) | 2018-01-23 |
| Publication | Publication Date | Title |
|---|---|---|
| CN107622239B (en) | A Method for Detection of Specified Building Areas in Remote Sensing Images Constrained by Hierarchical Local Structure | |
| CN111126202B (en) | Object detection method of optical remote sensing image based on hole feature pyramid network | |
| CN108549891B (en) | Multi-scale diffusion well-marked target detection method based on background Yu target priori | |
| CN109271991B (en) | License plate detection method based on deep learning | |
| CN112560675B (en) | Bird Vision Object Detection Method Combining YOLO and Rotation-Fusion Strategy | |
| CN108596055B (en) | Airport target detection method of high-resolution remote sensing image under complex background | |
| CN110309781B (en) | House damage remote sensing identification method based on multi-scale spectrum texture self-adaptive fusion | |
| CN105335966B (en) | Multiscale morphology image division method based on local homogeney index | |
| CN110136154A (en) | Semantic Segmentation Method of Remote Sensing Image Based on Fully Convolutional Network and Morphological Processing | |
| CN108596108B (en) | Aerial remote sensing image change detection method based on triple semantic relation learning | |
| CN110135354B (en) | Change detection method based on live-action three-dimensional model | |
| CN108629286B (en) | Remote sensing airport target detection method based on subjective perception significance model | |
| CN103632363A (en) | Object-level high-resolution remote sensing image change detection method based on multi-scale fusion | |
| CN109684922A (en) | A kind of recognition methods based on the multi-model of convolutional neural networks to finished product dish | |
| CN108305260A (en) | Detection method, device and the equipment of angle point in a kind of image | |
| CN110267101B (en) | Unmanned aerial vehicle aerial video automatic frame extraction method based on rapid three-dimensional jigsaw | |
| CN106557740B (en) | A Recognition Method of Oil Depot Targets in Remote Sensing Images | |
| CN111882573B (en) | A method and system for extracting cultivated land plots based on high-resolution image data | |
| CN115512247A (en) | Regional building damage grade assessment method based on image multi-parameter extraction | |
| CN116740758A (en) | Bird image recognition method and system for preventing misjudgment | |
| CN108446707A (en) | Remote sensing image airplane detection method based on key point screening and DPM confirmation | |
| CN111291818A (en) | A Sample Equalization Method for Cloud Mask-Oriented Non-Uniform Classes | |
| CN111241987A (en) | A Cost-Sensitive Three-Way Decision-Based Visual Tracking Method for Multi-target Models | |
| CN108133218A (en) | Infrared target detection method, equipment and medium | |
| Quispe et al. | Automatic building change detection on aerial images using convolutional neural networks and handcrafted features |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information | Inventor after:Bi Fukun Inventor after:Tian Yumeng Inventor after:Lei Mingyang Inventor after:Qin Yanyan Inventor after:Zhang Xu Inventor after:Zhang Feifei Inventor before:Bi Fukun Inventor before:Lei Mingyang Inventor before:Qin Yanyan Inventor before:Zhang Xu Inventor before:Zhang Feifei | |
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right | Effective date of registration:20200824 Address after:Room 801, 8 / F, 17 Shijingshan South Road, Shijingshan District, Beijing 100043 Patentee after:Beijing North Zhitu Information Technology Co., Ltd Address before:100144 Beijing City, Shijingshan District Jin Yuan Zhuang Road No. 5 Patentee before:North China University of Technology |