Warehouse identification method based on aerial image

Abstract

The invention discloses a warehouse identification method based on an aerial image, and relates to the technical field of image identification, and the method comprises the steps: obtaining an aerial image with a warehouse, and recognizing the basic features and auxiliary features of the warehouse; setting corresponding trust values for each basic feature and each auxiliary feature in advance; obtaining a to-be-recognized image, extracting data features of a building in the to-be-recognized image, and marking the data features as to-be-recognized features; comparing the to-be-identified features with basic features in a first basic database and auxiliary features in a first auxiliary database, and determining an identification result according to a comparison result; by combining the basic features and the auxiliary features and setting the trust value and the preset rule, the warehouse in the aerial image is identified more accurately, and the effect of improving the identification efficiency and accuracy is achieved; and by setting the trust value and the preset rule, the reliability and flexibility of identification are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to a warehouse recognition method based on aerial images. Background Art

[0002] Object recognition is to determine the properties of objects in an image based on the image taken by the camera. Object recognition can be used for map matching navigation, autonomous trajectory correction, etc., and has very important research significance.

[0003] In the fields of logistics, warehouse management, etc., accurate identification of warehouses is of great significance. Traditional warehouse identification methods mainly rely on manual inspections or ground photography, which are inefficient and greatly affected by environmental factors. With the development of drone technology and remote sensing technology, aerial images have become a new source of warehouse identification.

[0004] However, warehouse recognition in aerial images faces many challenges. In particular, in practical applications, misidentification often occurs, which seriously reduces the recognition accuracy. Summary of the invention

[0005] The present application solves the problem of the inability to accurately identify warehouses in the prior art by providing a warehouse identification method based on aerial images, thereby achieving the technical effect of improving the accuracy of warehouse identification.

[0006] The present application provides a warehouse identification method based on aerial images, the method comprising:

[0007] S100: Acquire an aerial image with a warehouse, identify basic features and auxiliary features of the warehouse, and store the aerial image with the basic features and the aerial image with the auxiliary features in a first basic database and a first auxiliary database respectively; set a corresponding trust value for each basic feature and auxiliary feature in advance; the trust value range is [0, 1];

[0008] S200: Acquire an image to be identified, extract data features of buildings in the image to be identified and mark them as features to be identified;

[0009] S300: Compare the feature to be identified with the basic feature in the first basic database and the auxiliary feature in the first auxiliary database, and determine the identification result according to the comparison result;

[0010] The basic features include the shape features, volume and area of ​​the building, and structural features of the building, and the auxiliary features include environmental features, text logo features, and supporting auxiliary features.

[0011] Furthermore, the feature to be identified is compared with the basic features in the first basic database and the auxiliary features in the first auxiliary database, respectively, including: calculating the similarity between the feature to be identified and each basic feature in the first basic database, marking the basic features whose similarity is greater than a preset first threshold as target basic features, calculating the number of the target basic features and the sum of the trust values ​​corresponding to all the target basic features, and obtaining a first number and a first trust value.

[0012] Further, determine whether the first number and the first trust value satisfy a first preset rule; if so, calculate the similarity between the feature to be identified and each auxiliary feature in the first auxiliary database, mark the auxiliary feature whose similarity is greater than a preset second threshold as a target auxiliary feature, calculate the sum of the trust values ​​corresponding to all the target auxiliary features, and obtain a second trust value; determine whether the second trust value satisfies a second preset rule; if so, determine that the feature to be identified is a warehouse, and use the sum of the first trust value and the second trust value as the total trust value of the feature to be identified; if the second preset rule is not satisfied, perform monitoring according to the long-term monitoring mechanism to obtain a monitoring result, and re-judge whether the second preset rule is satisfied based on the monitoring result.

[0013] Furthermore, the long-term monitoring mechanism is to continuously monitor the features to be identified within a preset time period, pre-set the shooting interval, take an image to be identified after each shooting interval, collect all the images to be identified within the preset time period, form a sequence of images, and analyze the dynamic change trajectory of the features to be identified through the sequence of images to obtain the monitoring results.

[0014] Further, if the first preset rule is not satisfied, the abnormality category is determined, and the abnormality category includes abnormality category one and abnormality category two; if it belongs to abnormality category one, a new feature to be identified of the image to be identified is obtained according to optimization mechanism one, and step S300 is repeated based on the new feature to be identified; if it belongs to abnormality category two, a new feature to be identified of the image to be identified is obtained according to optimization mechanism one, and step S300 is repeated based on the new feature to be identified and optimization mechanism two.

[0015] Furthermore, the optimization mechanism 1 is set to: extract edge information of the image to be identified, generate an edge image, divide the edge image into a preset number of cells, extract edge direction information for each cell, calculate the gradient direction of the edge pixel, quantize the gradient direction into several direction intervals, count the number of edge pixels in each direction interval, generate an edge distribution histogram, analyze the edge distribution histogram, and identify the edge direction feature as a new feature to be identified.

[0016] Furthermore, the optimization mechanism 2 is set to: superimpose any two images in the first basic database to form a second basic database; superimpose any two images in the first auxiliary database to form a second auxiliary database; and cross-superimpose any two images in the second basic database and the second auxiliary database to form an extended database.

[0017] Furthermore, the first preset rule is set as: the first number is not less than 3 and the first trust value is not less than a preset first trust threshold; the second preset rule is set as: the second trust value is not less than a preset second trust threshold.

[0018] Furthermore, the exception category one is set as: the first number is not less than 3 and the first trust value is less than a preset first trust threshold; the exception category two is set as: the first number is less than 3 and the first trust value is less than a preset first trust threshold.

[0019] Further, the method further includes: S400: obtaining a total trust value and an aerial image corresponding to the feature to be identified as a warehouse, identifying the corresponding background feature and marking it as a positive background feature;

[0020] Obtain the total trust value and aerial image corresponding to the unidentified feature identified as a non-warehouse, identify the corresponding background feature and mark it as a reverse background feature;

[0021] The occurrence frequencies of the positive background features and the negative background features in the corresponding aerial images are counted respectively, and the occurrence frequencies and the corresponding total trust values ​​are weightedly summed to obtain the added value;

[0022] The added value includes a positive added value and a negative added value. A positive added value is set for each positive background feature, and a negative added value is set for each negative background feature. The range of the positive added value is [0, 1], and the range of the negative added value is [-1, 0].

[0023] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0024] By combining basic features and auxiliary features, as well as setting trust values ​​and preset rules, warehouses in aerial images can be identified more accurately. A long-term monitoring mechanism is introduced to continuously monitor features to be identified that do not meet the conditions, reducing the possibility of misjudgment and missed judgments; achieving the effect of improving recognition efficiency and accuracy; and by setting trust values ​​and preset rules, the reliability and flexibility of recognition are enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 The figure is a schematic diagram of the overall process of a warehouse identification method based on aerial images in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] To facilitate the understanding of the present invention, the present application will be described more comprehensively below with reference to the relevant drawings; the drawings show preferred embodiments of the present invention, but the present invention can be implemented in many different forms and is not limited to the embodiments described herein; on the contrary, the purpose of providing these embodiments is to enable a more thorough and comprehensive understanding of the disclosed content of the present invention.

[0027] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which the present invention belongs; the terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention; the term "and / or" used herein includes any and all combinations of one or more related listed items.

[0028] Embodiment 1: Figure 1 As shown, a warehouse identification method based on aerial images, the method comprising:

[0029] S100: Acquire an aerial image with a warehouse, identify basic features and auxiliary features of the warehouse, and store the aerial image with the basic features and the aerial image with the auxiliary features in a first basic database and a first auxiliary database, respectively; set a corresponding trust value for each basic feature and auxiliary feature in advance, and the range of the trust value is [0, 1].

[0030] The basic features include the shape, volume and area of ​​the building, and the structural features of the building. The auxiliary features include environmental features, text mark features, and supporting auxiliary features. The environmental features include roads, trucks, and containers. The text mark refers to the name of the warehouse. There may be text marks such as "xx warehouse" on the top or outer wall of the warehouse. In aerial images, these text marks can be detected and used for warehouse identification through optical character recognition (OCR) technology. The supporting auxiliary features include supporting facilities such as parking lots, office areas, and fire protection facilities.

[0031] In some embodiments, a corresponding trust value is set for each basic feature and auxiliary feature. Different auxiliary features have different trust values ​​when combined with the extracted basic features. The subsequent recognition standard is determined according to the total trust value. When pre-setting the trust value, it is necessary to set different trust values ​​according to the frequency of occurrence of a certain feature in the aerial images of the warehouse. Preferably, the trust value of the basic feature is greater than that of the auxiliary feature. At the same time, different thresholds are also set according to the frequency of occurrence to distinguish whether the feature is a basic feature or an auxiliary feature. The specific setting is based on the actual situation, and this application does not impose any specific restrictions.

[0032] In some embodiments, the warehouse may be surrounded by supporting facilities such as parking lots, office areas, and firefighting facilities. These features may appear as obvious buildings, vehicles, or equipment in aerial images. Through image processing and computer vision technology, these features are detected and used for warehouse identification.

[0033] In some embodiments, a corresponding trust value is set in advance for each basic feature and auxiliary feature. The trust value is an indicator used to measure the credibility and importance of a feature in warehouse identification, and its value range is [0, 1]. The closer the trust value is to 1, the more reliable and important the feature is in warehouse identification; the closer the trust value is to 0, the less reliable or important the feature is in warehouse identification, which is reasonably determined based on the importance of the feature and the frequency of occurrence in the aerial image.

[0034] In some embodiments, when setting the trust value of basic features, the shape feature of the building is the key feature for identifying a warehouse, because the shape of a warehouse is usually relatively regular, such as a rectangle, square, etc. According to the frequency and accuracy of the shape feature in the aerial image, a higher trust value, such as 0.8, can be set for it; the volume and area of ​​the building are obtained through image measurement technology, which is particularly important for identifying large warehouses. Since there may be certain errors in the calculation of volume and area, but their contribution to warehouse identification is still large, a trust value slightly lower than the shape feature can be set for it, such as 0.75; the structural features of the building, such as the location of doors and windows, roof shape, etc. of the warehouse, are helpful for side identification of the warehouse. Since the structural features of different warehouses may vary, a moderate trust value, such as 0.6, can be set for it.

[0035] In some embodiments, when setting the trust value of auxiliary features, the road in the environmental feature is an important part of the warehouse environment, but its direct contribution to warehouse identification is limited, so a lower trust value can be set for it, such as 0.4; trucks and containers are usually related to the logistics activities of the warehouse, but not all trucks and containers appear around the warehouse, so their trust values ​​can be slightly higher than roads, such as 0.5; text logos such as warehouse names in text logo features are direct evidence for identifying warehouses and have high credibility. However, since the accuracy and reliability of OCR technology may be affected by many factors (such as image clarity, font type, etc.), a trust value slightly lower than that of basic features can be set for it, such as 0.7.

[0036] In some embodiments, when setting the trust value of supporting auxiliary features, a parking lot is a common supporting facility for a warehouse. Its presence has a certain auxiliary effect on identifying a warehouse, but not all warehouses are equipped with parking lots. Therefore, a moderate trust value can be set for it, such as 0.6; the office area is usually closely connected to the warehouse and is one of the important clues for identifying the warehouse. However, not all warehouses' office areas are clearly visible in aerial images, so a trust value slightly lower than that of the parking lot can be set for it, such as 0.55; firefighting facilities are crucial to the safety of the warehouse, but may not be easily perceived in aerial images. Therefore, its trust value can be relatively low, such as 0.5.

[0037] In some embodiments, the above-mentioned method of setting the trust value is only an example, and the specific setting needs to be made according to the actual situation and a large number of aerial images, and this application does not make any specific restrictions.

[0038] S200: Acquire an image to be identified, extract data features of buildings in the image to be identified, and mark them as features to be identified.

[0039] In some embodiments, the image to be identified is subjected to preprocessing operations such as geometric correction, radiation correction, and denoising to improve the quality and recognition of the image; an edge detection algorithm is used to extract the outline and shape features of the building in the image to be identified, image segmentation technology is used to separate the building from the background, and its area and perimeter are calculated, and computer vision technology is used to detect the structural features of the building, such as doorways, loading and unloading platforms, etc.; edge detection and linear feature extraction algorithms are used to detect the existence and direction of roads; image segmentation and object recognition technology are used to detect the existence and number of trucks and containers; image processing and computer vision technology are used to detect supporting facilities such as parking lots, office areas, and fire-fighting facilities; extracting features from images is a prior art and will not be elaborated in this application.

[0040] S300: Compare the feature to be identified with the basic feature in the first basic database and the auxiliary feature in the first auxiliary database respectively, and determine the identification result according to the comparison result.

[0041] In some embodiments, the buildings in the aerial images are first preliminarily screened based on basic features to exclude buildings that do not meet warehouse characteristics; then the preliminarily screened buildings are re-identified and a comprehensive judgment is made based on the number and type of auxiliary features. If there are many roads, trucks and containers around the building, or there are obvious text signs and supporting facilities, it is more likely to be identified as a warehouse.

[0042] The feature to be identified is compared with the basic features in the first basic database and the auxiliary features in the first auxiliary database respectively, including: calculating the similarity between the feature to be identified and each basic feature in the first basic database, marking the basic features whose similarity is greater than a preset first threshold as target basic features, calculating the number of the target basic features and the sum of the trust values ​​corresponding to all the target basic features, and obtaining a first number and a first trust value.

[0043] In some embodiments, first, data features of the building (i.e., features to be identified) are extracted from the image to be identified, including shape features, volume and area, structural features, etc. of the building. At the same time, representations of each basic feature are extracted from the first basic database, which are pre-stored for comparison with the features to be identified. A similarity measurement method (such as Euclidean distance, cosine similarity, shape context matching, etc.) is used to calculate the similarity between the feature to be identified and each basic feature in the first basic database. The choice of similarity measurement method depends on the type and representation of the feature. For example, for shape features, shape context or Fourier descriptors can be used for matching; for volume and area, the numerical difference can be directly calculated.

[0044] In some embodiments, a first threshold is pre-set to determine whether the similarity between the feature to be identified and the basic feature is high enough to consider them as a match, and the similarity between the feature to be identified and each basic feature is compared with the first threshold. If the similarity is greater than the first threshold, the basic feature is marked as a target basic feature, and the number of features marked as target basic features is counted to obtain a first number. This number reflects the degree of match between the building in the image to be identified and the basic features in the first basic database. For each basic feature marked as a target basic feature, its corresponding trust value is searched, and the trust values ​​of all target basic features are added to obtain a first trust value. This value reflects the overall credibility and importance of the building in the image to be identified in terms of the basic features.

[0045] Determine whether the first number and the first trust value satisfy a first preset rule; if so, calculate the similarity between the feature to be identified and each auxiliary feature in the first auxiliary database, mark the auxiliary feature whose similarity is greater than a preset second threshold as a target auxiliary feature, calculate the sum of the trust values ​​corresponding to all the target auxiliary features, and obtain a second trust value; determine whether the second trust value satisfies a second preset rule; if so, determine that the feature to be identified is a warehouse, and use the sum of the first trust value and the second trust value as the total trust value of the feature to be identified; if the second preset rule is not satisfied, perform monitoring according to the long-term monitoring mechanism to obtain a monitoring result, and re-determine whether the second preset rule is satisfied according to the monitoring result.

[0046] In some embodiments, the first preset rule is set as follows: the first number is not less than 3 and the first trust value is not less than a preset first trust threshold; the second preset rule is set as follows: the second trust value is not less than a preset second trust threshold.

[0047] In some embodiments, for each auxiliary feature in the first auxiliary database, calculate its similarity with the feature to be identified, mark the auxiliary features whose similarity is greater than a preset second threshold as target auxiliary features, calculate the sum of the trust values ​​corresponding to all target auxiliary features, obtain the second trust value, check whether the second trust value is not less than the preset second trust threshold, and if the second trust value satisfies the second preset rule (i.e., not less than the second trust threshold), determine that the feature to be identified is a warehouse. The sum of the first trust value and the second trust value is used as the total trust value of the feature to be identified for subsequent evaluation or recording. If the second trust value does not meet the second preset rule, the long-term monitoring mechanism is started.

[0048] For example, if the first number and the first trust value of the features to be identified satisfy the first preset rule, it indicates that there are many valuable similar basic features, and then the auxiliary features are compared. If there are text identification features and corresponding vehicles and containers in the auxiliary features, it can be determined that the features to be identified are warehouses; if the corresponding auxiliary features are not detected, or only the remaining auxiliary features of lower value are detected, further judgment is required.

[0049] In some embodiments, the long-term monitoring mechanism is to continuously monitor the features to be identified within a preset time period, pre-set shooting intervals, take an image to be identified after each shooting interval, collect all the images to be identified within the preset time period, form a sequence of images, and analyze the dynamic change trajectory of the features to be identified through the sequence of images to obtain monitoring results.

[0050] In some embodiments, the long-term monitoring mechanism continuously monitors changes in the feature to be identified and its surrounding environment over a period of time. The similarity between the feature to be identified and the basic feature and the auxiliary feature, as well as the corresponding trust value, can be recalculated regularly or irregularly, and based on the monitoring results obtained by the long-term monitoring mechanism, it is re-determined whether the second trust value meets the second preset rule. If during the monitoring period, the second trust value meets the requirements of the second preset rule, it can be determined that the feature to be identified is a warehouse.

[0051] In some embodiments, when shooting in the area to be monitored, it is necessary to ensure that the critical areas in the warehouse and the surrounding areas that may change are covered. At the same time, it should have a timed shooting or triggered shooting function, support high-resolution image capture, to adapt to different lighting and weather conditions, and pre-set the time period and shooting interval according to the regularity of warehouse activities (such as daily receipt and delivery time), such as shooting once every hour, every half day or a specific time period, shooting continuously for 10 or 15 days, etc.; ensure that the shooting frequency is high enough to capture the dynamic changes of the image activities to be identified, but also consider storage and processing capabilities to avoid data overload. Use cloud storage or local servers to store the captured images, establish an image database, organize the image data in chronological order, form a sequence of images, attach metadata such as timestamps and shooting locations to each image, implement image quality inspection, eliminate unqualified images such as blur, overexposure or underexposure, and ensure the quality of the sequence images.

[0052] In some embodiments, it is necessary to analyze the dynamic change trajectory of the features to be identified with respect to the sequence images to obtain monitoring results, and perform analysis based on the required data status of the actual warehouse. If the required data status of the actual warehouse is met, it is determined to be a warehouse; if not, it is determined not to be a warehouse; the required data status of the actual warehouse includes the entry and exit of goods, changes in the number of containers, personnel activities and the use of equipment. Specifically, the frequency, time and loading status of trucks entering and leaving the warehouse are monitored; by observing changes in the stacking areas inside or outside the warehouse, it is inferred whether there is stored goods and the quantity of stored goods; the activity patterns of staff inside and outside the warehouse are recorded, such as loading and unloading, inspections, etc.; and the use of logistics equipment such as forklifts and cranes is monitored.

[0053] In some embodiments, the monitoring method of sequential images can use computer vision technology to compare the sequential images frame by frame, identify the position changes of goods, vehicles, personnel or equipment, analyze the activity patterns of personnel or vehicles through machine learning algorithms, and identify regular behaviors, such as scheduled receipt and delivery of goods; extract key features (such as truck specifications, cargo types, and cargo quantities), and track changes in these features in sequential images to monitor inventory and logistics dynamics; the monitoring method of sequential images is a technical means well known to people in this field, and this application does not impose specific limitations.

[0054] In some embodiments, the extracted features to be identified are first compared and identified with the basic features in the first basic database, and a judgment is made by calculating the similarity between the two features. A first threshold is set in advance, and the basic features with a similarity greater than the first threshold are marked as target basic features. The number of the target basic features is calculated, and the sum of the trust values ​​corresponding to all the target basic features is calculated to obtain a first trust value.

[0055] In some embodiments, when calculating the sum of the corresponding trust values ​​for the features to be identified, it is necessary to make a comprehensive judgment based on the trust values ​​of the corresponding basic features themselves and the similarities between them. For example, the trust values ​​of basic features 1, 2, and 3 are 0.6, 0.8, and 0.7, respectively. The similarities between the features to be identified and basic features 1, 2, and 3 are 80%, 90%, and 78%, respectively. The corresponding trust value that should be assigned to the features to be identified is 0.6×80%+0.8×90%+0.7×78%=1.746, which is the sum of the trust values ​​obtained.

[0056] In some embodiments, a warehouse recognition model is constructed based on the first basic database and the first auxiliary database; the image to be recognized is collected by aerial photography using high-altitude flying tools such as drones, and first a large number of aerial images are collected, including images containing warehouses and images without warehouses; the warehouses in the images are manually annotated to generate a training data set, and a suitable deep learning model, such as a convolutional neural network (CNN), is selected as the basis for image recognition; the model is trained using the training data set, and the parameters of the model are adjusted to improve the accuracy and efficiency of recognition, and the trained model is optimized, such as adjusting the network structure and using data enhancement technology to improve the generalization ability of the model, and the model is tested using a test data set to evaluate its performance; the trained model is deployed to actual application scenarios, such as drone aerial image recognition systems. The aerial image to be recognized is input, and the model automatically outputs the recognition result and marks the warehouse location in the image.

[0057] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0058] This application more accurately identifies warehouses in aerial images by combining basic features and auxiliary features, and setting trust values ​​and preset rules. It introduces a long-term monitoring mechanism to continuously monitor the features to be identified that do not meet the conditions, thereby reducing the possibility of misjudgment and missed judgments; it achieves the effect of improving recognition efficiency and accuracy; and by setting trust values ​​and preset rules, it enhances the reliability and flexibility of recognition.

[0059] Embodiment 2: This embodiment is a further improvement on the above embodiment.

[0060] If the first preset rule is not satisfied, the abnormal category is determined, and the abnormal category includes abnormal category one and abnormal category two; if it belongs to abnormal category one, a new feature to be identified of the image to be identified is obtained according to optimization mechanism one, and step S300 is repeated based on the new feature to be identified; if it belongs to abnormal category two, a new feature to be identified of the image to be identified is obtained according to optimization mechanism one, and step S300 is repeated based on the new feature to be identified and optimization mechanism two.

[0061] In some embodiments, the exception category one is set as: the first number is not less than 3 and the first trust value is less than a preset first trust threshold; the exception category two is set as: the first number is less than 3 and the first trust value is less than a preset first trust threshold.

[0062] The optimization mechanism 1 is set as follows: extract edge information of the image to be identified, generate an edge image, divide the edge image into a preset number of cells, extract edge direction information for each cell, calculate the gradient direction of the edge pixel, quantize the gradient direction into a number of direction intervals, count the number of edge pixels in each direction interval, generate an edge distribution histogram, analyze the edge distribution histogram, and identify edge direction features as new features to be identified.

[0063] In some embodiments, firstly, image enhancement techniques such as histogram equalization, adaptive histogram equalization (CLAHE), sharpening filter, etc. are used on the image to be identified in abnormal category 1 to improve the contrast and clarity of the image, and the enhanced image is denoised to reduce the interference of noise on subsequent feature extraction; basic features with low confidence values ​​are further analyzed, such as shape features, volume and area features, structural features, etc. An edge extraction algorithm (such as the Canny operator method) is used to obtain the edge information of the image and generate an edge image.

[0064] In some embodiments, the edge image is divided into a preset number of cells (such as an 8x8 or 16x16 grid), the edge direction information is extracted for each cell, the gradient direction of the edge pixel is calculated, the gradient direction is quantified into several direction intervals (such as (0, 45), (45, 90), (90, 135), (135, 180) degrees, etc.), the number of edge pixels in each direction interval is counted, and an edge distribution histogram is generated; the data enhancement and edge extraction mentioned above belong to the prior art, and this application does not go into details. The edge distribution histogram is analyzed to identify the main edge direction features, and the identified edge direction features are matched and verified in combination with the feature information in the first basic database. According to the matching results, the basic feature trust value in the feature to be identified is adjusted, or new basic features are added.

[0065] In some embodiments, based on the refined features and the adjusted trust values, the image to be identified is compared with the first basic database and the first auxiliary database again. A new similarity, a target basic feature quantity, a first trust value, and a target auxiliary feature quantity and a second trust value are calculated. It is determined whether the new first quantity and the first trust value meet the first preset rule, and whether the new second quantity and the second trust value meet the second preset rule. If the rules are met, the recognition result is output; if still not met, the features of the image recognition are further adjusted.

[0066] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0067] This application introduces exception category one and exception category two to classify and process situations that do not meet the preset rules; uses edge extraction algorithm and edge distribution histogram analysis to extract edge direction features as new features to be identified; for images of exception category one, image enhancement technology and denoising processing are used to improve image quality and the accuracy of feature extraction; further analyzes basic features with low trust values, matches and verifies them in combination with feature information in the database, adjusts feature trust values ​​and supplements new features, and achieves the accuracy, reliability and stability of the recognition warehouse.

[0068] Embodiment 3: This embodiment makes further improvements on the basis of the above contents.

[0069] The second optimization mechanism is set as follows: any two images in the first basic database are superimposed to form a second basic database; any two images in the first auxiliary database are superimposed to form a second auxiliary database; any two images in the second basic database and the second auxiliary database are cross-superimposed to form an extended database.

[0070] In some embodiments, any two images in the first basic database are superimposed to remove duplicate features, retain and integrate unique features to form a second basic database; any two images in the first auxiliary database are superimposed to integrate environmental features, text identification features and supporting auxiliary features to enhance the diversity of auxiliary features to form a second auxiliary database; any two images in the second basic database and the second auxiliary database are cross-superimposed to form new scene feature sets, which include a combination of basic features and auxiliary features to simulate the complexity of the actual warehouse environment, and finally form an extended database.

[0071] In some embodiments, there are multiple technical methods for superimposing images in the database, and features in different databases are integrated into a unified data structure using fusion techniques such as feature vector splicing and feature space mapping. Algorithms (such as K-means clustering and PCA dimensionality reduction) are used to remove duplicate or redundant features, optimize feature sets, and improve database efficiency. Based on the integrated feature sets, virtual warehouse scenes are constructed. These scenes contain a combination of multiple basic features and auxiliary features for simulating complex and changeable warehouse environments. This is achieved through existing technologies and is not specifically limited in this application.

[0072] In some embodiments, the extended database will contain a richer set of features, including single features, combined features, and simulated scene features. These features are stored in a structured form to facilitate rapid retrieval and matching. By extending the database, the scope of feature matching can be expanded, and the ability to recognize uncommon or complex features in the image to be identified can be improved, thereby effectively solving the problem of abnormal category two.

[0073] In some embodiments, if it belongs to abnormal category two, new features to be identified of the image to be identified are obtained according to optimization mechanism one, feature recognition is performed based on the new features to be identified and the extended database generated by optimization mechanism two, and step S300 is repeated; the identified edge direction features are matched and verified with the features in the extended database, and according to the matching results, the basic feature trust values ​​in the features to be identified are adjusted, or new basic features are supplemented to enhance the integrity of the feature set.

[0074] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0075] This application constructs a second basic database, a second auxiliary database, and a second basic database through image superposition and feature integration technology, thereby forming an extended database, enriching the feature set, improving the diversity and complexity of the database, realizing feature fusion and optimization, removing duplicate or redundant features, and improving the efficiency of the database; based on the integrated feature set, a virtual warehouse scene is constructed to simulate the complexity of the actual warehouse environment, thereby improving the adaptability of the recognition system to complex environments, and enhancing the adaptability and robustness of the recognition system to complex environments.

[0076] Embodiment 4: This embodiment makes further improvements on the basis of the above contents.

[0077] The method also includes: S400: obtaining the total trust value and aerial image corresponding to the feature to be identified as a warehouse, identifying the corresponding background feature and marking it as a positive background feature; obtaining the total trust value and aerial image corresponding to the feature to be identified as a non-warehouse, identifying the corresponding background feature and marking it as a negative background feature; respectively counting the occurrence frequency of the positive background feature and the negative background feature in the corresponding aerial image, and performing weighted summation of the occurrence frequency and the corresponding total trust value to obtain an added value.

[0078] Specifically, the added value includes a positive added value and a negative added value. A positive added value is set for each positive background feature, and a negative added value is set for each negative background feature. The range of the positive added value is [0, 1], and the range of the negative added value is [-1, 0].

[0079] In some embodiments, in the recognition process, in addition to extracting the positive features (basic features and auxiliary features) of the building, background features are also extracted. Background features include but are not limited to: the surrounding environment (such as natural landscapes, urban buildings, industrial areas, etc.), features of non-warehouse buildings, specific geographical markers (such as rivers, mountains, road network patterns), etc. The background features of known warehouse and non-warehouse aerial images are labeled and classified to form a background feature library. An additional value is set for each background feature based on the frequency of occurrence of the background feature in the warehouse and non-warehouse images and the corresponding total trust value and the specificity of the background. Some background features (such as industrial areas) may have a positive indicative effect on warehouse identification, while others (such as natural landscapes) may have a negative indicative effect. The specificity of the background refers to having obvious background features that are very easy to identify.

[0080] In some embodiments, when the positive feature recognition result is unclear in step S300 (such as the first number or the first trust value is close to the threshold), the background feature trust value is introduced as an auxiliary judgment basis, and a new decision rule is set to comprehensively consider the trust values ​​of the positive features and background features to make a final recognition judgment, thereby further improving the accuracy of warehouse recognition based on aerial images.

[0081] In this embodiment, the extraction and utilization of background features are introduced. By identifying and analyzing the background features in aerial images, an additional source of information is provided for warehouse identification. According to the frequency of occurrence of background features in warehouse and non-warehouse images and the corresponding total trust value, an additional value is set for each background feature, and the contribution of background features to warehouse identification is quantified. When the positive feature recognition result is unclear, the background feature trust value is introduced as an auxiliary judgment basis, and a new decision rule is set to improve the accuracy of recognition.

[0082] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0083] This application provides more information sources and judgment bases for warehouse identification by introducing background features and their added values, thereby improving the accuracy of identification. When the positive feature identification result is unclear, the background features can be used for auxiliary judgment, thereby enhancing the system's adaptability and robustness to complex environments. When the positive feature identification result is close to the threshold, the background feature trust value is introduced as an auxiliary judgment basis, which effectively solves the problem of unclear positive feature identification and further improves the reliability and stability of identification.

[0084] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A warehouse identification method based on aerial images, characterized in that: The method comprises: S100: Acquire an aerial image with a warehouse, identify basic features and auxiliary features of the warehouse, and store the aerial image with the basic features and the aerial image with the auxiliary features in a first basic database and a first auxiliary database respectively; set a corresponding trust value for each basic feature and auxiliary feature in advance; the trust value range is [0, 1]; S200: Acquire an image to be identified, extract data features of buildings in the image to be identified and mark them as features to be identified; S300: Compare the features to be identified with the basic features in the first basic database and the auxiliary features in the first auxiliary database, and determine the identification results according to the comparison results; The basic features include the shape features, volume and area of ​​the building, and structural features of the building, and the auxiliary features include environmental features, text logo features, and supporting auxiliary features.

2. A warehouse identification method based on aerial images as claimed in claim 1, characterized in that: The feature to be identified is compared with the basic features in the first basic database and the auxiliary features in the first auxiliary database respectively, including: calculating the similarity between the feature to be identified and each basic feature in the first basic database, marking the basic features whose similarity is greater than a preset first threshold as target basic features, calculating the number of the target basic features and the sum of the trust values ​​corresponding to all the target basic features, and obtaining a first number and a first trust value.

3. A warehouse identification method based on aerial images as claimed in claim 2, characterized in that: Determine whether the first number and the first trust value satisfy a first preset rule; if so, calculate the similarity between the feature to be identified and each auxiliary feature in the first auxiliary database, mark the auxiliary feature whose similarity is greater than a preset second threshold as a target auxiliary feature, calculate the sum of the trust values ​​corresponding to all the target auxiliary features, and obtain a second trust value; determine whether the second trust value satisfies a second preset rule; if so, determine that the feature to be identified is a warehouse, and use the sum of the first trust value and the second trust value as the total trust value of the feature to be identified; If the second preset rule is not satisfied, monitoring is performed according to the long-term monitoring mechanism to obtain a monitoring result, and whether the second preset rule is satisfied is re-determined according to the monitoring result.

4. A warehouse identification method based on aerial images as claimed in claim 3, characterized in that: The long-term monitoring mechanism is to continuously monitor the features to be identified within a preset time period, pre-set shooting intervals, take a picture of the features to be identified after each shooting interval, collect all the images to be identified within the preset time period, form a sequence of images, and analyze the dynamic change trajectory of the features to be identified through the sequence of images to obtain the monitoring results.

5. The warehouse identification method based on aerial images as claimed in claim 3, characterized in that: If the first preset rule is not satisfied, the abnormal category is determined, and the abnormal category includes abnormal category one and abnormal category two; if it belongs to abnormal category one, a new feature to be identified of the image to be identified is obtained according to optimization mechanism one, and step S300 is repeated based on the new feature to be identified; if it belongs to abnormal category two, a new feature to be identified of the image to be identified is obtained according to optimization mechanism one, and step S300 is repeated based on the new feature to be identified and optimization mechanism two.

6. A warehouse identification method based on aerial images as claimed in claim 5, characterized in that: The optimization mechanism 1 is set as follows: extract edge information of the image to be identified, generate an edge image, divide the edge image into a preset number of cells, extract edge direction information for each cell, calculate the gradient direction of the edge pixel, quantize the gradient direction into a number of direction intervals, count the number of edge pixels in each direction interval, generate an edge distribution histogram, analyze the edge distribution histogram, and identify edge direction features as new features to be identified.

7. The warehouse identification method based on aerial images as claimed in claim 5, characterized in that: The second optimization mechanism is set as follows: any two images in the first basic database are superimposed to form a second basic database; any two images in the first auxiliary database are superimposed to form a second auxiliary database; Any two images in the second basic database and the second auxiliary database are cross-superimposed to form an extended database.

8. The warehouse identification method based on aerial images as claimed in claim 3, characterized in that: The first preset rule is set as: the first number is not less than 3 and the first trust value is not less than a preset first trust threshold; The second preset rule is set as: the second trust value is not less than a preset second trust threshold.

9. The warehouse identification method based on aerial images as claimed in claim 5, characterized in that: The exception category one is set as: the first number is not less than 3 and the first trust value is less than a preset first trust threshold; the exception category two is set as: the first number is less than 3 and the first trust value is less than a preset first trust threshold.

10. The warehouse identification method based on aerial images according to claim 1, characterized in that: The method further includes: S400: obtaining a total trust value and an aerial image corresponding to the feature to be identified as a warehouse, identifying the corresponding background feature and marking it as a positive background feature; Obtain the total trust value and aerial image corresponding to the unidentified feature identified as a non-warehouse, identify the corresponding background feature and mark it as a reverse background feature; The occurrence frequencies of the positive background features and the negative background features in the corresponding aerial images are counted respectively, and the occurrence frequencies and the corresponding total trust values ​​are weightedly summed to obtain the added value; The added value includes a positive added value and a negative added value. A positive added value is set for each positive background feature, and a negative added value is set for each negative background feature. The range of the positive added value is [0, 1], and the range of the negative added value is [-1, 0].

Images