Building facade defect detection system based on UAV external thermal excitation compensation

Abstract

The present invention belongs to the field of data analysis and processing, and discloses a building facade defect detection system based on unmanned aerial vehicle exogenous thermal excitation compensation; the system includes generating a three-dimensional model of the building facade, predefining a heterogeneous modal joint optimization framework, automatically connecting ports according to mode selection, and running an intelligent control center; deploying a drone formation based on a master-slave model, and the formation sharing detection data through federated learning; the intelligent control center includes deploying an onboard lightweight model, processing thermal imaging data and RGB images in real time on the drone end, identifying suspected defect areas, generating a spot distribution thermal map through a pre-inspection model deployed by the intelligent control center, and performing heating tasks in a dynamic light-heat collaborative integration; the intelligent control center also includes ground analysis and centralized control, using a cloud-based actuarial model to construct a spatiotemporal map of heat diffusion and a prediction of hollowing expansion trends, and finally generating a three-dimensional defect distribution map and displaying it through an interactive interface, thereby realizing drone defect detection of building facades under exogenous thermal excitation compensation.

Description

Technical Field

[0001] The present invention relates to the field of data analysis and processing, and more particularly, to a building facade defect detection system based on unmanned aerial vehicle excitation compensation. Background Art

[0002] With the acceleration of urbanization, defects such as hollowing and gapping in the exterior walls of high-rise buildings are becoming increasingly prominent. Traditional manual inspection methods (such as Spider-Man tapping inspection) have shortcomings such as low efficiency, high risk, and strong subjectivity. In recent years, drone thermal imaging inspection technology has gradually been applied to building facade defect detection, but the following technical bottlenecks still exist:

[0003] Existing technologies often rely on natural temperature differences (such as sunlight) or ambient temperature fluctuations for thermal imaging inspection, limiting detection efficiency depending on environmental conditions. In low-temperature environments or on complex building facades (such as areas with dense decorative moldings), thermal radiation signals are weak and chaotically distributed, making it difficult to accurately identify defects. Furthermore, existing solutions lack the ability to detect deep defects (>5cm) and cannot meet high-precision requirements.

[0004] While a few studies have attempted to incorporate active heat sources, their heating modes are limited and unable to dynamically adapt to varying building materials and geometric features. The fixed heating spot shape and unadjustable power waste energy and can easily damage building surfaces. Existing technologies have yet to address the integration of laser / microwave synergistic heating, dynamic spot control, and multimodal feedback, and their ability to characterize complex defects (such as hollowing accompanied by cracks) is insufficient. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned objectives, the present invention provides the following technical solution: a building facade defect detection system based on drone external thermal excitation compensation, comprising: generating a three-dimensional building facade model;

[0006] Input the 3D building facade model, predefine the heterogeneous modal joint optimization framework, automatically connect ports based on the mode selection, and run the intelligent control center;

[0007] Based on the master-slave deployment, drone formations fly along inspection paths optimized by a dynamic task allocation algorithm and share inspection data through federated learning.

[0008] The intelligent control center deploys an airborne lightweight model, processes thermal imaging data and RGB images in real time on the drone side, identifies suspected defect areas, and marks local locations requiring further inspection, all of which are recorded as target areas. The loaded drone hovers over the target area, generates a thermal map of the light spot distribution using the pre-inspection model deployed by the intelligent control center, activates the heat source excitation device, and performs the heating task through dynamic light-heat synergy integration.

[0009] The pre-inspection model uses RGB images of the building facade's three-dimensional model surface to generate spot distribution through real-time analysis using a Transformer-based neural network. It then integrates with the digital twin library and a meta-learning adaptation engine to quickly match heating strategies.

[0010] After the heat source excitation device is activated, the liquid metal droplet array is controlled by the electrowetting effect to dynamically adjust the spot shape and size. The spatiotemporal modulation algorithm alternately implements the laser-microwave synergistic heating strategy and simultaneously activates the acoustic-thermal resonance enhancement mechanism.

[0011] The intelligent control center also includes ground analysis and centralized control, which uses cloud-based actuarial models to construct spatiotemporal maps of heat diffusion and predict hollowing expansion trends, ultimately generating a three-dimensional defect distribution map and displaying it through an interactive interface.

[0012] Preferably, the heterogeneous modal joint optimization framework supports switching between three operation modes, including manual remote operation mode, human-machine collaborative decision-making mode and intelligent autonomous operation mode;

[0013] Manual remote operation mode means that the operator manually controls the drone to fly to the inspection area and perform the heating task through the remote control.

[0014] The human-machine collaborative decision-making mode means that the operator pre-sets the inspection area and automatically plans the drone's inspection path and performs the heating task.

[0015] The intelligent autonomous operation mode means automatically planning the drone inspection path based on the three-dimensional model of the building facade, independently designing the inspection area and executing the heating task, for fully automated inspection.

[0016] Preferably, the three-dimensional model of the building facade is generated by BIM or point cloud data, and multispectral SLAM is used to fuse visible light, thermal infrared and the three-dimensional model of the building facade to construct a 3D semantic map, in which the geometric features and thermal physical parameters of the building surface are marked;

[0017] Based on the 3D model of the building facade, a drone formation is designed, and the formation flies along an inspection path optimized by a dynamic task allocation algorithm.

[0018] Preferably, the method for detecting path flight based on dynamic task allocation algorithm optimization includes:

[0019] The dynamic task allocation algorithm is based on the distributed decision-making mechanism of auction theory;

[0020] The distributed decision-making mechanism models the heat source excitation, data collection, and path planning involved in a detection mission as an auction market, with drones acting as bidders and the mission as a commodity. Each drone calculates its bid value based on its own capabilities and mission requirements, taking into account payload, endurance, and sensor type. The drone with the highest bid value is then assigned to it.

[0021] Preferably, the drone formation deployment is based on a master-slave drone formation, using one heavy-duty drone and three light-weight detection drones to form a heterogeneous cluster, covering the building facade according to the detection path, and sharing the detection data through federated learning;

[0022] Among them, the heavy-duty UAV is equipped with a heat source excitation device, and the light detection aircraft includes an infrared thermal imager, millimeter wave radar and ultrasonic sensor;

[0023] The heavy-duty drone uses an infrared thermal imager to conduct a preliminary scan of the building facade, recording the wall temperature distribution, while the lightweight detector simultaneously collects multimodal data, including thermal images, millimeter wave data, ultrasonic data, and RGB images;

[0024] An airborne lightweight model is deployed to process thermal imaging data and RGB images in real time on the drone side, identify suspected defect areas, and mark local locations that require further inspection. These are all recorded as target areas. The determination of suspected defect areas is based on temperature difference anomaly thresholds and geometric feature anomaly thresholds. The thresholds are adaptively adjusted according to environmental conditions and material properties.

[0025] Preferably, the load-carrying drone hovers over the target area, generates a light spot distribution thermal map through a pre-inspection model deployed by the intelligent control center, activates the heat source excitation device, and performs the heating task through dynamic light-heat synergy integration, including:

[0026] The heat source type is automatically selected using a meta-learning adaptation engine based on the characteristics of the target area and detection requirements;

[0027] Dynamic light-heat synergy integration works alternately through a spatiotemporal modulation algorithm. During the heating process, the intelligent control center adjusts the heating parameters through a dynamic thermal compensation decision tree.

[0028] The acoustic and thermal resonance enhancement mechanism is activated simultaneously, and the piezoelectric ultrasonic transducer array applies mechanical vibrations of a specific frequency to amplify the temperature difference signal in the hollow area;

[0029] After heating is completed, the heat source excitation device is turned off, and the light detection aircraft in the drone formation uses an infrared thermal imager to continuously record the temporal changes in temperature differences in the target area. The specific duration is adaptively adjusted according to the wall material and ambient temperature. The millimeter wave radar and ultrasonic sensor synchronously collect deep defect and surface crack data, and the RGB camera collects geometric feature data to obtain a comprehensive data package.

[0030] Preferably, the method of generating a spot distribution heat map using a pre-check model deployed by an intelligent control center includes:

[0031] A pre-built digital twin library includes cross-scale thermal properties of building materials, embedded with material microstructure scanning data and heat conduction simulation models. Heat source parameters are initialized based on environmental parameters, and a meta-learning adaptation engine and a dynamic thermal compensation decision tree are added. Meta-learning uses small sample learning to quickly match parameters, while the dynamic thermal compensation decision tree automatically adjusts heating parameters based on material type, environmental conditions, and geometric characteristics.

[0032] Based on the surface RGB image of the three-dimensional model of the building facade, the pre-inspection model uses a neural network built with the Transformer architecture to analyze it and generate a spot distribution heat map in real time to guide the laser-microwave collaborative heating strategy.

[0033] Preferably, the method for performing a heating task through dynamic light-heat synergistic integration includes:

[0034] Dynamic light-heat synergy integration is based on the laser-microwave synergistic heating strategy, using the electrowetting effect to control the liquid metal droplet array, thereby changing the spot shape and size;

[0035] Among them, controlling the liquid metal droplet array through the electrowetting effect includes a lens array driven by a micro-electrode grid, changing the surface tension of the liquid metal droplets by applying an electric field, and dynamically adjusting the light spot;

[0036] The spatiotemporal modulation algorithm involves designing a pulse modulation controller that alternates between 532nm laser for surface heating and 24GHz microwaves for deep penetration. The laser pulse period is 50ms (ON / OFF) and the microwave pulse period is 100ms. An infrared thermal imager records the temporal changes in temperature during the heating and cooling processes, generating thermal imaging time series data. Subsequently, an inversion algorithm is used to calculate the material's layered thermal conductivity.

[0037] The acoustic-thermal resonance enhancement mechanism includes installing a transducer array below the drone's heat source module, adaptively adjusting the operating frequency based on the wall material and defect type, customizing the transducer power range, and regulating the vibration amplitude through closed-loop control.

[0038] Preferably, the airborne lightweight model adopts edge computing technology and deploys a lightweight neural network to perform real-time coarse defect positioning;

[0039] The rough defect positioning is transmitted to the ground analysis and control center in real time through the high-speed wireless communication module for detailed regional analysis;

[0040] Ground analysis and centralized control utilizes cloud-based actuarial models to fuse and analyze multimodal data from the comprehensive data package of the coarse positioning area. Graph neural networks construct spatiotemporal maps of heat diffusion to accurately locate and quantify voids, hollows, and cracks. Physical information neural networks predict the expansion trend of hollow areas.

[0041] Among them, the thermal diffusion spatiotemporal graph uses graph neural networks to model the spatiotemporal relationship of thermal diffusion. Combined with the constraints of the heat conduction equation, the graph neural network models the thermal field changes through nodes and edges;

[0042] Physical information neural network integrates heat conduction partial differential equation constraints to predict the expansion trend of hollowing area.

[0043] Preferably, the three-dimensional defect distribution map is generated based on the building facade point cloud data, and displayed through an interactive interface, and supports accurate marking of the defect location, which is displayed through a detailed inspection report, including the 3D coordinates of the defect location, defect type, area, depth, temperature difference change curve, expansion trend prediction, and overlay display of visible light, thermal imaging, millimeter wave data and 3D point cloud, and supports output in multiple formats.

[0044] The technical effects and advantages of the building facade defect detection system based on UAV external thermal excitation compensation of the present invention are as follows:

[0045] 1. Detection accuracy and depth are significantly improved

[0046] Active thermal excitation compensation technology was designed. Through laser-microwave synergistic heating (532nm laser surface heating and 24GHz microwave deep penetration) and acoustic-thermal resonance enhancement (20-100kHz mechanical vibration), the temperature difference signal in the hollowing area was significantly amplified, achieving a breakthrough in recognition accuracy.

[0047] Using a liquid metal droplet array controlled by the electrowetting effect, dynamic adjustment of the spot shape and size can be achieved in milliseconds, accurately matching complex geometric features such as architectural decorative lines and curtain wall joints, and improving defect positioning accuracy.

[0048] 2. Enhanced environmental adaptability

[0049] Combining infrared thermal imaging, millimeter-wave radar, ultrasonic sensors, and RGB camera data, we build a thermal-acoustic-mechanical multi-physics coupling analysis model to maintain detection stability in extreme environments such as strong sunlight, rain, and fog.

[0050] The dynamic thermal compensation decision tree adjusts the heating power and duration in real time based on the material type, ambient temperature and humidity to avoid overheating or insufficient energy.

[0051] 3. High intelligence and high precision

[0052] The three-level operation mode switches seamlessly, participates in the distributed task allocation algorithm to shorten the task allocation response time, improves the efficiency of heterogeneous UAV formation collaboration, models the spatiotemporal relationship of heat conduction based on the graph neural network (GNN), and combines the physical information neural network (PINN) to integrate the heat conduction partial differential equation constraints to realize the four-dimensional (3D space + time) dynamic evolution prediction of the hollow area, and greatly reduces the prediction error. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 Schematic diagram of the steps of the building facade defect detection system based on UAV external thermal excitation compensation of the present invention;

[0054] Figure 2 Schematic diagram of the processing flow of the airborne lightweight model in the present invention;

[0055] Figure 3 This is a schematic diagram of the processing flow of ground analysis and centralized control in the present invention. DETAILED DESCRIPTION

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0057] Example 1

[0058] See also Figure 1 、 Figure 2 and Figure 3 As shown, the building facade defect detection system based on drone external thermal excitation compensation described in this embodiment includes:

[0059] This design aims to address the limitations of traditional drone thermal imaging inspection in low-temperature environments, complex lines, openings, and walls made of different materials. By introducing exogenous thermal excitation compensation technology, combined with drone platforms, multimodal sensor fusion, adaptive optical systems, and intelligent algorithms, a set of efficient, safe, and accurate building facade defect detection technologies is formed. The innovative goals include:

[0060] 1. Intelligent light-heat synergy: Through a bionic adaptive optical engine and multimodal thermal field modulation technology, dynamic optimization of light spot shape and heat source distribution is achieved to adapt to complex building facade features.

[0061] 2. Material Adaptive Thermal Management: Build a digital twin library of building surfaces, using meta-learning and dynamic thermal compensation algorithms to adapt to different wall materials and environmental conditions.

[0062] 3. Edge Intelligent Analysis: Design a layered AI processing system that combines edge computing and cloud actuarial analysis to achieve real-time defect identification and evolution prediction.

[0063] 4. Autonomous operation: Improve detection efficiency and safety through multi-agent collaborative strategies and environmental adaptive navigation.

[0064] Specific content includes:

[0065] The heterogeneous modal joint optimization framework supports switching between three operating modes, including manual remote control mode, human-machine collaborative decision-making mode, and intelligent autonomous operation mode;

[0066] Manual remote control operation mode means that the operator manually controls the drone to fly to the inspection area and perform the heating task through the remote control.

[0067] The human-machine collaborative decision-making mode means that the operator (based on experience or prejudgment) pre-sets the inspection area and automatically plans the drone's inspection path and performs the heating task.

[0068] The intelligent autonomous operation mode means automatically planning the drone inspection path based on the three-dimensional model of the building facade, independently designing the inspection area and executing the heating task, for fully automated inspection.

[0069] The 3D building facade model is generated using BIM or point cloud data. Multispectral SLAM is used to fuse visible light, thermal infrared, and the 3D building facade model to construct a 3D semantic map with thermal property labels. The map is marked with geometric features of the building surface (such as line width, curvature, and opening shape) and thermal property parameters (such as thermal conductivity and reflectivity). The map supports real-time updates and adapts to dynamic environmental changes (such as occlusion and wind speed).

[0070] Based on the 3D model of the building facade, a drone formation is designed, and the formation flies along an inspection path optimized by a dynamic task allocation algorithm.

[0071] The method for detecting path flight based on dynamic task allocation algorithm optimization includes:

[0072] The dynamic task allocation algorithm is based on a distributed decision-making mechanism based on auction theory, optimizing flight paths and sensor scheduling strategies in real time.

[0073] The distributed decision-making mechanism models the heat source activation, data collection, and path planning components of an inspection task as an auction market, with drones acting as "bidders" and the task as the "commodity." Each drone evaluates its own capabilities (calculated using a weighted algorithm) based on payload, range, and sensor type, as well as task requirements (distance and priority), and calculates its bid value (also calculated using a weighted algorithm). The drone with the highest bid value is then assigned. For example, heavy-duty drones are prioritized for heat source activation tasks, while light-duty drones are prioritized for data collection. This dynamic task allocation algorithm can reduce task conflicts by approximately 30% and optimize overall inspection efficiency.

[0074] The drone formation deployment is based on a master-slave drone formation, using one heavy-duty drone and three light-weight detection drones to form a heterogeneous cluster. The drones cover the building facades according to the detection path and share the detection data through federated learning.

[0075] Among them, the heavy-duty UAV is equipped with a heat source excitation device, and the light detection aircraft includes an infrared thermal imager, millimeter wave radar and ultrasonic sensor;

[0076] The heavy-duty drone uses an infrared thermal imager to conduct a preliminary scan of the building facade, recording the wall temperature distribution. The lightweight detector simultaneously collects multimodal data, including thermal images (temperature difference distribution), millimeter wave data (deep defects), ultrasonic data (surface cracks), and RGB images (geometric features).

[0077] An airborne lightweight model is deployed to process thermal imaging data and RGB images in real time on the drone side, identify suspected defect areas (such as areas with abnormal temperature differences and abnormal lines), and mark local locations that require further inspection. These are all recorded as target areas. The judgment of suspected defect areas is based on the temperature abnormality threshold and the geometric feature abnormality threshold. The threshold is adaptively adjusted according to environmental conditions and material properties.

[0078] The formation structure is automatically adjusted based on mission requirements. For example, in large-area wall inspection tasks, one heavy-duty drone (heat source excitation) and three lightweight detectors (multi-sensor) form a "star formation" to improve coverage efficiency. In complex line area inspection tasks, the formation is reorganized into a "chain formation" to achieve precise scanning section by section.

[0079] The following examples illustrate how the determination of suspected defect areas is based on temperature difference anomaly thresholds and geometric feature anomaly thresholds, with the thresholds adaptively adjusted according to environmental conditions and material properties:

[0080] Assume that a building facade detection scenario is as follows:

[0081] Ambient temperature: 25℃;

[0082] Materials: Common exterior wall materials (such as concrete, masonry, etc.);

[0083] Detection targets: wall cracks, peeling, holes and deep defects;

[0084] Temperature difference abnormal threshold setting, calculation formula:

[0085] Temperature difference abnormal threshold = basic temperature difference + ambient temperature compensation coefficient × (current ambient temperature - standard ambient temperature);

[0086] Example:

[0087] The current ambient temperature is 25°C, and the standard ambient temperature is 20°C.

[0088] The ambient temperature compensation coefficient is 0.1℃ / ℃.

[0089] The basic temperature difference was set at 5°C.

[0090] The temperature difference abnormal threshold is: 5℃+0.1℃ / ℃×(25℃-20℃) = 5℃+0.5℃=5.5℃.

[0091] Calculation formula for setting geometric feature anomaly threshold:

[0092] Geometric feature anomaly threshold = basic geometric feature value × material property coefficient;

[0093] Example:

[0094] For concrete walls, basic geometric characteristic values ​​(such as crack width) are set to 0.1 mm.

[0095] The material property coefficient is 1.5 (determined according to the properties of concrete).

[0096] The geometric feature anomaly threshold is: 0.1mm×1.5=0.15mm.

[0097] The load-carrying drone hovers over the target area, generates a light spot distribution thermal map using the pre-inspection model deployed by the intelligent control center, activates the heat source excitation device, and performs the heating task through dynamic light-heat synergy integration. The method includes:

[0098] The heat source type (laser, microwave, infrared lamp) is automatically selected using a meta-learning adaptation engine based on the characteristics of the target area and detection requirements. For example, 532nm laser is preferred for line detection, 24GHz microwave is preferred for deep void detection, and infrared lamp is used for large-area preheating.

[0099] Dynamic light-heat synergy and integration work alternately through a spatiotemporal modulation algorithm. During the heating process, the intelligent control center adjusts the heating parameters (power, time, and spot shape) through a dynamic thermal compensation decision tree. For example, for glass curtain walls with low thermal conductivity, pulse mode heating is adopted, the power curve is a square wave (300W constant), and the monitoring frequency is 100Hz; for concrete walls with high thermal conductivity, a continuous scanning mode is adopted, the power curve is an exponential increase (200W→500W), and the monitoring frequency is 10Hz.

[0100] The acoustic and thermal resonance enhancement mechanism is activated simultaneously, and the piezoelectric ultrasonic transducer array applies mechanical vibration of a specific frequency to amplify the temperature difference signal in the hollowing area. For example, for concrete walls, low-frequency vibration (20kHz) is used to enhance deep hollowing signals; for glass curtain walls, high-frequency vibration (100kHz) is used to enhance surface crack signals.

[0101] After heating is completed, the heat source excitation device is turned off, and the light detection aircraft in the drone formation uses an infrared thermal imager to continuously record the temporal changes in temperature differences in the target area (cooling process) for 3 to 10 minutes. The specific duration is adaptively adjusted according to the wall material and ambient temperature. The millimeter-wave radar and ultrasonic sensor synchronously collect deep defect and surface crack data, and the RGB camera collects geometric feature data to obtain a comprehensive data package.

[0102] The comprehensive data package can include millimeter-wave radar and ultrasonic sensors synchronously collecting deep defect and surface crack data, RGB camera collecting geometric feature data, compressed thermal imaging images, RGB images, millimeter-wave data, ultrasonic data, 3D point cloud data / 3D semantic maps, and environmental parameters. The specific content can be adaptively selected.

[0103] The weight of the heat source excitation device is controlled within 5.5 kilograms, the power supply system supports hybrid power supply mode, and the load interface supports modular replacement.

[0104] Methods for generating spot distribution heat maps using the pre-inspection model deployed in the intelligent control center include:

[0105] A pre-built digital twin library includes cross-scale thermal properties of building materials (including newly added metal curtain walls, wood, and composite panels), embeds material microstructure scanning data and heat conduction simulation models, and initializes heat source parameters (type, power, spot shape, and size) based on environmental parameters (temperature, humidity, wind speed, and light intensity). A meta-learning adaptation engine and a dynamic thermal compensation decision tree are also added. Meta-learning uses small-sample learning to quickly match parameters. The meta-learning adaptation engine quickly adapts to new materials and scenarios without extensive retraining. The dynamic thermal compensation decision tree automatically adjusts heating parameters based on material type, environmental conditions, and geometric characteristics.

[0106] For example, in constructing a digital twin library, a cross-scale thermophysical property database encompasses 15 types of building materials (including metal curtain walls, wood, and composite panels, which will be added dynamically based on market conditions). This database includes parameters such as thermal conductivity, thermal diffusivity, surface reflectivity, microporosity, and thermal expansion coefficient. This database incorporates material microstructure scan data obtained using a scanning electron microscope and a thermal conduction simulation model, based on the finite element method, that simulates the temperature differentials experienced by different materials during heating and cooling. The database supports real-time query and updates to accommodate new materials and scenarios. For example, for glass curtain walls, the database records their low thermal conductivity and high reflectivity; for stone wall surfaces, its high porosity and thermal expansion coefficient are recorded.

[0107] The digital twin library can improve the material parameter matching accuracy by about 25%, providing reliable data support for the dynamic thermal compensation algorithm.

[0108] In addition, for example, for new materials, the meta-learning adaptation engine achieves fast parameter matching of new materials through small sample learning, including:

[0109] The meta-learning model utilizes the MAML (Model-Agnostic Meta-Learning) framework. The pre-trained model is based on data from 15 material categories in the digital twin library. Simply inputting a small number of new material samples (5-10) allows for rapid adaptation of heating parameters. Model outputs include heat source type, power curve, heating time, and spot shape. For example, for a newly introduced composite panel material, the system determined the optimal heating mode to be a pulsed laser, a square wave power curve, and a 50Hz monitoring frequency based on a small amount of experimental data.

[0110] Compared with traditional machine learning models, the meta-learning adaptation engine can reduce the new material adaptation time by about 80% and improve parameter matching accuracy by about 15%.

[0111] The dynamic thermal compensation algorithm based on the decision tree automatically adjusts the heating mode, power curve, and monitoring frequency according to the material type, environmental conditions, and geometric characteristics. The decision tree input parameters include material type, ambient temperature, humidity, wind speed, line width, and curvature. The output parameters include heat source type, heating mode, power curve, and monitoring frequency. A partial example of the decision tree:

[0112] The concrete heating mode is continuous scanning, the power curve is exponential increase (200W→500W), and the monitoring frequency is 10Hz;

[0113] The glass curtain wall's heating mode is pulsed (50ms on / off), the power curve is square wave (300W constant), and the monitoring frequency is 100Hz. A dynamic thermal compensation decision tree improves heating efficiency by approximately 20% and ensures accurate detection of wall surfaces of varying materials.

[0114] The "Dynamic Thermal Compensation Algorithm" automatically adjusts heat source power and heating time based on ambient temperature, humidity, wind speed, and wall material. For example, in low-temperature, high-humidity environments, it automatically increases heating power (by 20%) and extends heating time (by 30%) to ensure that the wall temperature rise meets detection requirements.

[0115] Based on the surface RGB image of the building facade 3D model (which can also be a pre-scanned RGB image of the building surface), the pre-inspection model uses a neural network built with a Transformer architecture to analyze it and generate a spot distribution heat map in real time to guide the laser-microwave collaborative heating strategy; that is, the heating task is used to guide the dynamic focusing strategy of the laser beam, including the adjustment of the laser beam spot shape and size, and the microwave energy output.

[0116] The neural network built using the Transformer architecture consists of a visual encoder, a temporal decoder, and a heat map generation layer. The input is an RGB image of the building's surface, and the output is a heat map of the spot distribution, which guides the focus position and shape of the laser beam. The model training data includes simulated images and measured data of various building lines, openings, and materials. The predictor optimizes the spot distribution and improves heating efficiency by analyzing the geometric features (such as line width and curvature) and material properties (such as reflectivity) of the building surface. Compared to traditional fixed spot strategies, the deep learning beam predictor can improve heating efficiency by approximately 30% and significantly enhance detection accuracy in complex line areas.

[0117] The "laser + microwave" multi-mode heat source excitation solution uses AI algorithms to achieve dynamic switching of heat source modes. Laser heat sources are suitable for high-precision local heating, while microwave heat sources are suitable for large-area deep heating. The two modes can be automatically switched or used in combination according to detection requirements. Compared with single heat source excitation, multi-mode heat source excitation significantly improves the adaptability of the system and can cope with the detection needs of different wall materials (such as concrete, ceramic tiles, glass curtain walls, etc.), different defect types (such as surface voids, deep hollowing) and different environmental conditions (such as low temperature, high humidity, strong wind, etc.). Existing patented technologies mostly use a single heat source (such as laser or microwave) and lack the ability to work in multiple modes in collaboration. This solution fills the technical gap of multi-mode heat source excitation through heat source switching algorithms and optical system optimization.

[0118] Methods for performing heating tasks through dynamic light-heat synergy integration include:

[0119] Dynamic light-heat synergy is based on a laser-microwave synergistic heating strategy (by matching the complex geometric features of the building facade in a 3D model (such as European-style carvings and curtain wall joints)). The electrowetting effect is used to control the liquid metal droplet array (material: gallium-indium alloy), thereby changing the shape (circular, linear, annular) and size of the light spot (in milliseconds).

[0120] The electrowetting effect is used to control the liquid metal droplet array, including a lens array driven by a micro-electrode grid. By applying an electric field, the surface tension of the liquid metal droplets is changed, dynamically adjusting the light spot. The light spot adjustment range is 0.02m² to 1.5m², with a response time of ≤10ms. The light spot shape is automatically adjusted by using the RGB image of the building surface and 3D point cloud data in the real-time three-dimensional building facade model. For example, for fine lines less than 0.1m in width, the light spot is adjusted to a linear shape; for large wall surfaces, the light spot is adjusted to a rectangular shape. Compared to traditional mechanical zoom lenses, liquid metal lens arrays are approximately 50% lighter, consume approximately 40% less power, and can adapt to complex lines and curved walls.

[0121] The spatiotemporal modulation algorithm involves designing a pulse modulation controller that alternates between 532nm laser for surface heating and 24GHz microwaves for deep penetration. The laser pulse period is 50ms (ON / OFF) and the microwave pulse period is 100ms. An infrared thermal imager records the temporal changes in temperature during the heating and cooling processes, generating thermal imaging time series data. An inversion algorithm is then used to calculate the material's layered thermal conductivity (surface layer, base layer, and adhesive layer).

[0122] The inversion algorithm is based on the heat conduction equation, the heat source distribution function is modeled using the spatiotemporal distribution of laser and microwaves, and the thermal diffusion coefficient and heat source absorption coefficient are obtained through experimental calibration. Laser-microwave synergistic heating can increase the detection depth of deep defects to 10 cm (approximately 50% higher than a single heat source) and enhance the inspection accuracy of complex wall surfaces.

[0123] The acoustic-thermal resonance enhancement mechanism involves mounting a transducer array beneath the drone's heat source module. The operating frequency is adaptively adjusted based on the wall material and defect type. For example, for concrete walls, low-frequency vibration (20kHz) enhances deep hollowing signals; for glass curtain walls, high-frequency vibration (100kHz) enhances surface crack signals. The transducer power range is customizable, for example, from 10W to 50W, and the vibration amplitude is adjusted through closed-loop control to ensure no damage to the wall. Acoustic-thermal resonance technology can increase the temperature difference signal strength in the hollowing area by approximately 20%, significantly improving detection sensitivity.

[0124] The airborne lightweight model uses edge computing technology and deploys a lightweight neural network to achieve real-time coarse defect positioning;

[0125] For example, a lightweight model based on TinyML technology compresses the defect detection network to 2MB and runs on the drone's embedded processor. The model inputs are thermal images and RGB images, and outputs coarse localization of suspected defect areas, with latency controlled to less than 50ms. Model training data includes data on various wall materials, environmental conditions, and other factors. The lightweight model primarily performs real-time coarse localization, identifying suspected defect areas (such as areas with abnormal temperature differences and line irregularities) and marking locations requiring further inspection. The model is optimized through pruning and quantization techniques to ensure efficient operation on low-power embedded processors.

[0126] The coarse defect positioning is transmitted to the ground analysis system in real time through a high-speed wireless communication module for detailed analysis of the area; the airborne lightweight model significantly reduces the amount of data transmission through the edge computing framework, and the delay is controlled within 50ms.

[0127] The ground analysis system uses cloud-based actuarial models to fuse and analyze the multimodal data in the comprehensive data package of the coarse positioning area. The graph neural network constructs a spatiotemporal map of heat diffusion, accurately locates and quantifies voids, hollows, and cracks, and the physical information neural network predicts the expansion trend of the hollow area; for example, it generates short-term (within 1 year) and long-term (within 5 years) evolution prediction results.

[0128] The analysis system leverages a multimodal data fusion framework to accurately locate defects such as voids, hollows, and cracks, and classify and quantify the defect type and severity. For example, the system can output void area (in m²), hollow depth (in mm), crack width (in mm), and thermal conductivity anomaly coefficient (calculated based on temporal changes in temperature differentials).

[0129] The cloud-based actuarial model is designed based on a 3D thermal field analysis engine using graph neural networks (GNNs). The input is multimodal data (thermal images, RGB images, 3D point cloud data, and environmental parameters) in a comprehensive data package, and the output is the quantitative results of defect location, type, and severity.

[0130] Among them, the thermal diffusion spatiotemporal map uses graph neural networks to model the spatiotemporal relationship of thermal diffusion. Combined with the constraints of the heat conduction equation, the graph neural network models the thermal field changes through nodes (representing local areas of the wall) and edges (representing thermal diffusion relationships); accurately locates defects such as voids, hollows and cracks. Compared with traditional convolutional neural networks, graph neural networks can improve the accuracy of complex thermal field analysis by about 20%, and are especially suitable for defect detection on multi-layer material walls.

[0131] The heat conduction equation constraint can be expressed as ,in, is the temperature field, is the position and time function. is the thermal diffusivity, which indicates how quickly heat spreads inside an object. is the Laplace operator, which represents the spatial variation of the temperature field. is the heat source absorption coefficient, which indicates the degree of influence of the heat source on the temperature field. is the heat source distribution function, which represents the distribution of the heat source in space and time. It should be determined according to the actual heat source distribution. For example, if the heat source is a point heat source, it can be represented by the Dirac function.

[0132] A physical information neural network incorporates constraints from the partial differential equation for heat conduction to predict the expansion trend of hollowing areas. For example, the physical information neural network combines data-driven and physical constraints. Its inputs are thermal imaging time-series data, 3D point cloud data, and environmental parameters, and its output is the expansion trend of the hollowing area (area growth rate, depth change rate). The model is trained by minimizing a physical loss function (based on the heat conduction equation) and a data loss function (based on measured data). Predictions include short-term (within one year) and long-term (within five years) trends in hollowing expansion, providing a basis for maintenance decisions. The defect evolution prediction module can improve the scientific nature of reform decisions, reduce unnecessary maintenance costs, and extend the service life of buildings.

[0133] The system generates real-time temperature difference distribution maps, 3D defect markers, and quantified results of suspected defect areas for operator reference. For complex lines or openings, the system automatically triggers in-depth analysis mode, using high-precision models to further confirm the defect type and severity.

[0134] A three-dimensional defect distribution map is generated based on the building facade point cloud data and displayed through an interactive interface. It also supports accurate marking of the defect location. The accurate marking of the defect location is displayed through a detailed inspection report, including the 3D coordinates of the defect location, defect type, area, depth, temperature difference change curve, expansion trend prediction, and overlay display of visible light, thermal imaging, millimeter wave data and 3D point cloud. It supports output in multiple formats (such as PDF, CAD, BIM model) to facilitate subsequent maintenance and file management.

[0135] The inspection report also includes repair recommendations. For example, it recommends prioritizing repairs for severely hollowed areas, regular monitoring for minor hollowing, and structural reinforcement for areas with rapidly expanding cracks. Precise annotations with an accuracy of ±5cm are included, and the inspection report is automatically generated in PDF format, including defect type, location, area, severity, and repair recommendations.

[0136] For identified high-risk defect areas, the system supports automatic review function, and the drone returns to the designated area for secondary heat source excitation and thermal imaging detection to ensure the accuracy of the detection results.

[0137] The intelligent control center integrates a "heat source overload protection algorithm" to monitor the temperature and power of the heat source module in real time, and automatically reduces the power or shuts down the heat source when the safety threshold is exceeded.

[0138] The drone has added "multi-sensor obstacle avoidance sensing", integrating lidar, ultrasonic sensor and infrared sensor, supporting 360° obstacle avoidance, ensuring safe flight in complex building facade environments.

[0139] Data backup and recovery supports local storage and cloud backup of detection data. It automatically switches to local storage mode when communication is interrupted, and automatically uploads data to the cloud after the task is completed.

[0140] Example 2

[0141] See also Figure 1 、 Figure 2 and Figure 3 As shown, for parts not described in detail in this embodiment, please refer to the description of Example 1. A building facade defect detection system based on drone exogenous thermal excitation compensation is provided, including:

[0142] The operator inputs the 3D model of the building facade or manually sets the inspection area, and the system automatically plans the drone's flight path and heat source excitation plan.

[0143] After the drone takes off, it collects data such as external temperature, humidity, and wind speed in real time, and the dynamic thermal compensation algorithm adjusts the heat source parameters based on the environmental data.

[0144] The drone is equipped with a thermal imager to conduct a preliminary scan of the building facade, generate an initial thermal image, and identify potential defective areas (such as lines and areas with abnormal local temperature differences).

[0145] For areas where initial thermal imaging cannot accurately identify voids or defects, the system automatically activates the heat source activation mechanism. Based on the wall material, environmental conditions, and defect type, it dynamically selects a laser or microwave heat source, and uses the optical system to adjust the spot size and irradiation position. The heat source irradiation time is generally 3-5 minutes, causing the surface temperature of the target area to transiently rise to approximately 40°C.

[0146] After the heat source is activated, the heat source device is turned off, and the thermal imager immediately begins recording the temporal changes in the temperature difference in the target area. The recording time is generally 5 to 10 minutes, with a capture frequency of 1 frame per second, ensuring that the complete cooling process is captured.

[0147] Data transmission and analysis:

[0148] Thermal imaging data, environmental data, and drone status data are transmitted in real time to the ground control station via a 5G communication module. Analysis system software processes the data to analyze defect type, location, and severity.

[0149] The system generates a 3D defect distribution map and inspection report, including defect type (such as voids, hollows, cracks, etc.), location coordinates, area, severity, and repair recommendations. Inspection results can be viewed in real time on the ground control station display and can be stored and shared in the cloud.

[0150] In addition, the original design can be supplemented according to the diverse scene requirements during use, including:

[0151] The liquid metal lens array utilizes gallium-indium alloy droplets, achieving millisecond-level dynamic transformation of light spot morphology through the electrowetting effect (response time <10ms). Multi-level spot control technology can also be added, dividing the lens array into four independently controlled zones to simultaneously generate multiple spot morphologies (e.g., a circular spot in the center for localized heating, and a linear spot at the edge for line scanning). The resolution of spot morphology transformation has been increased to 0.1mm, enabling precise matching of complex geometric features such as European-style carvings and curtain wall joints.

[0152] In order to improve the heating efficiency of laser on highly reflective materials such as glass curtain walls, a "polarization dynamic adjustment module" can be added to adjust the polarization direction of the laser in real time through a liquid crystal polarizer (range 0°~90°), reducing the reflectivity to below 5% and increasing the thermal energy utilization rate by 20%.

[0153] The spatiotemporal modulation algorithm can optionally incorporate a "thermal field uniformity optimization mechanism" and "microwave phase modulation technology." The "thermal field uniformity optimization mechanism" dynamically adjusts the laser and microwave pulse widths (within a 10ms range of 90%) by monitoring thermal imaging data in real time, ensuring uniform temperature rise across the target area (temperature difference <2°C). The "microwave phase modulation technology" generates a controllable microwave interference field using a multi-antenna array, focusing microwave energy at a specific depth within the wall (within a range of 1cm to 10cm), enabling accurate detection of deep hollowing.

[0154] The piezoelectric ultrasonic transducer array can also be upgraded to a "wideband adaptive transducer," supporting dynamic adjustment of the frequency range from 10kHz to 200kHz. It generates acoustic wave focus through phase delay control, enhancing the thermoelastic effect in the hollow area (amplifying the temperature difference signal by a factor of 3). To reduce the interference of acoustic waves on the drone's stability, a new "acoustic isolation cover" has been added, made of porous sound-absorbing material and weighing less than 0.5 kg.

[0155] To further amplify the thermal response differences in defective areas, alternating hot and cold excitation can be used. After heat source excitation, the target area is rapidly cooled using a jet-type liquid nitrogen cooler (at a cooling rate of 10°C / s). The temporal changes in temperature differences under hot and cold cycles are recorded to improve the accuracy of identifying deep cracks (identification depth increased to 12 cm).

[0156] Example 3

[0157] This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the building facade defect detection system based on drone exogenous thermal excitation compensation provided above is implemented.

[0158] Since the electronic device introduced in this embodiment is the electronic device used to implement the building facade defect detection system based on drone external thermal excitation compensation in the embodiment of this application, based on the building facade defect detection system based on drone external thermal excitation compensation introduced in the embodiment of this application, those skilled in the art can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application will not be described in detail here. As long as those skilled in the art implement the electronic device used in the building facade defect detection system based on drone external thermal excitation compensation in the embodiment of this application, it falls within the scope of protection of this application.

[0159] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0160] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that for users of ordinary skill in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A building facade defect detection system based on UAV external thermal excitation compensation is characterized by: Including generating 3D model of building facade: Input the 3D building facade model, predefine the heterogeneous modal joint optimization framework, automatically connect ports based on the mode selection, and run the intelligent control center; Based on the master-slave deployment, drone formations fly along inspection paths optimized by a dynamic task allocation algorithm and share inspection data through federated learning. The intelligent control center deploys an airborne lightweight model, processes thermal imaging data and RGB images in real time on the drone side, identifies suspected defect areas, and marks local locations requiring further inspection, all of which are recorded as target areas. The loaded drone hovers over the target area, generates a thermal map of the light spot distribution using the pre-inspection model deployed by the intelligent control center, activates the heat source excitation device, and performs the heating task through dynamic light-heat synergy integration. The pre-inspection model uses RGB images of the building facade's three-dimensional model surface to generate spot distribution through real-time analysis using a Transformer-based neural network. This model then uses a digital twin library and a meta-learning adaptation engine to quickly match heating strategies. After the heat source excitation device is activated, the electrowetting effect is used to control the liquid metal droplet array to dynamically adjust the spot shape and size. The spatiotemporal modulation algorithm alternately implements the laser-microwave synergistic heating strategy and simultaneously activates the acoustic-thermal resonance enhancement mechanism. The intelligent control center also includes ground analysis and centralized control, which uses cloud-based actuarial models to construct spatiotemporal maps of heat diffusion and predict hollowing expansion trends, ultimately generating a three-dimensional defect distribution map and displaying it through an interactive interface.

2. The building facade defect detection system based on UAV external thermal excitation compensation according to claim 1 is characterized in that: The heterogeneous modal joint optimization framework supports three operation mode switches, including manual remote operation mode, human-machine collaborative decision-making mode and intelligent autonomous operation mode; Manual remote control mode means that the operator manually controls the drone to fly to the inspection area and perform the heating task through the remote control; The human-machine collaborative decision-making mode means that the operator pre-sets the inspection area, and the drone automatically plans the inspection path and performs the heating task; The intelligent autonomous operation mode means automatically planning the drone inspection path based on the three-dimensional model of the building facade, independently designing the inspection area and executing the heating task, for fully automated inspection.

3. The building facade defect detection system based on UAV external thermal excitation compensation according to claim 2 is characterized in that: The three-dimensional model of the building facade is generated by BIM or point cloud data, and multispectral SLAM is used to fuse visible light, thermal infrared and the three-dimensional model of the building facade to construct a 3D semantic map, in which the geometric features and thermal physical properties of the building surface are marked; Based on the 3D model of the building facade, a drone formation is designed, and the formation flies along an inspection path optimized by a dynamic task allocation algorithm.

4. The building facade defect detection system based on UAV external thermal excitation compensation according to claim 3 is characterized in that: The method for detecting path flight based on dynamic task allocation algorithm optimization includes: The dynamic task allocation algorithm is based on the distributed decision-making mechanism of auction theory; The distributed decision-making mechanism models the heat source excitation, data collection, and path planning in the detection mission as an auction market, with drones as bidders and missions as commodities. Each drone evaluates its own capabilities and mission requirements based on payload, endurance, and sensor type, and calculates the bid value. The drone with the highest bid value is allocated to it.

5. The building facade defect detection system based on UAV external thermal excitation compensation according to claim 4 is characterized in that: The drone formation deployment is based on a master-slave drone formation, using one heavy-duty drone and three light-weight detection drones to form a heterogeneous cluster, covering the building facade according to the detection path and sharing the detection data through federated learning; Among them, the heavy-duty UAV is equipped with a heat source excitation device, and the light detection aircraft includes an infrared thermal imager, millimeter wave radar and ultrasonic sensor; The heavy-duty drone uses an infrared thermal imager to conduct a preliminary scan of the building facade, recording the wall temperature distribution, while the lightweight detector simultaneously collects multimodal data, including thermal images, millimeter wave data, ultrasonic data, and RGB images; An airborne lightweight model is deployed to process thermal imaging data and RGB images in real time on the drone side, identify suspected defect areas, and mark local locations that require further inspection. These are all recorded as target areas. The determination of suspected defect areas is based on temperature difference anomaly thresholds and geometric feature anomaly thresholds. The thresholds are adaptively adjusted according to environmental conditions and material properties.

6. The building facade defect detection system based on UAV external thermal excitation compensation according to claim 5 is characterized in that: The load-carrying drone hovers over the target area, generates a light spot distribution thermal map through a pre-inspection model deployed by the intelligent control center, activates a heat source excitation device, and performs a heating task through dynamic light-heat synergy integration. The method includes: The heat source type is automatically selected using a meta-learning adaptation engine based on the characteristics of the target area and detection requirements; Dynamic light-heat synergy integration works alternately through a spatiotemporal regulation algorithm. During the heating process, the intelligent control center adjusts the heating parameters through a dynamic thermal compensation decision tree. The acoustic and thermal resonance enhancement mechanism is activated simultaneously, and the piezoelectric ultrasonic transducer array applies mechanical vibrations of a specific frequency to amplify the temperature difference signal in the hollow area; After heating is completed, the heat source excitation device is turned off, and the light detection aircraft in the drone formation uses an infrared thermal imager to continuously record the temporal changes in temperature differences in the target area. The specific duration is adaptively adjusted according to the wall material and ambient temperature. The millimeter wave radar and ultrasonic sensor synchronously collect deep defect and surface crack data, and the RGB camera collects geometric feature data to obtain a comprehensive data package.

7. The building facade defect detection system based on UAV external thermal excitation compensation according to claim 6 is characterized in that: The method for generating a spot distribution heat map using a pre-inspection model deployed by the intelligent control center includes: A pre-built digital twin library includes cross-scale thermal properties of building materials, embedded with material microstructure scanning data and heat conduction simulation models. Heat source parameters are initialized based on environmental parameters, and a meta-learning adaptation engine and a dynamic thermal compensation decision tree are added. Meta-learning uses small sample learning to quickly match parameters, while the dynamic thermal compensation decision tree automatically adjusts heating parameters based on material type, environmental conditions, and geometric characteristics. Based on the surface RGB image of the three-dimensional model of the building facade, the pre-inspection model uses a neural network built with the Transformer architecture to analyze it and generate a spot distribution heat map in real time to guide the laser-microwave collaborative heating strategy.

8. The building facade defect detection system based on UAV external thermal excitation compensation according to claim 7 is characterized in that: The method for performing a heating task through dynamic light-heat synergistic integration includes: Dynamic light-heat synergy integration is based on the laser-microwave synergistic heating strategy, using the electrowetting effect to control the liquid metal droplet array, thereby changing the spot shape and size; Among them, controlling the liquid metal droplet array through the electrowetting effect includes a lens array driven by a micro-electrode grid, changing the surface tension of the liquid metal droplets by applying an electric field, and dynamically adjusting the light spot; The spatiotemporal modulation algorithm involves designing a pulse modulation controller that alternates between 532nm lasers for surface heating and 24GHz microwaves for deep penetration. The laser pulse period is 50ms, and the microwave pulse period is 100ms. An infrared thermal imager records the temporal changes in temperature during the heating and cooling processes, generating thermal imaging time series data. Subsequently, an inversion algorithm is used to calculate the material's layered thermal conductivity. The acoustic-thermal resonance enhancement mechanism includes installing a transducer array below the drone's heat source module, adaptively adjusting the operating frequency based on the wall material and defect type, customizing the transducer power range, and regulating the vibration amplitude through closed-loop control.

9. The building facade defect detection system based on UAV external thermal excitation compensation according to claim 8 is characterized in that: The airborne lightweight model uses edge computing technology and deploys a lightweight neural network to perform real-time coarse defect positioning; The rough defect positioning is transmitted to the ground analysis and control center in real time through the high-speed wireless communication module for detailed regional analysis; Ground analysis and centralized control utilizes cloud-based actuarial models to fuse and analyze multimodal data from the comprehensive data package of the coarse positioning area. Graph neural networks construct spatiotemporal maps of heat diffusion to accurately locate and quantify voids, hollows, and cracks. Physical information neural networks predict the expansion trend of hollow areas. Among them, the thermal diffusion spatiotemporal graph uses graph neural networks to model the spatiotemporal relationship of thermal diffusion. Combined with the constraints of the heat conduction equation, the graph neural network models the thermal field changes through nodes and edges; Physical information neural network integrates heat conduction partial differential equation constraints to predict the expansion trend of hollowing area.

10. The building facade defect detection system based on UAV external thermal excitation compensation according to claim 9 is characterized in that: The three-dimensional defect distribution map is generated based on the building facade point cloud data and displayed through an interactive interface. It supports accurate marking of the defect location and is displayed through a detailed inspection report, including the 3D coordinates of the defect location, defect type, area, depth, temperature difference change curve, expansion trend prediction, and overlay display of visible light, thermal imaging, millimeter wave data and 3D point cloud, and supports output in multiple formats.

Images