Photovoltaic construction unmanned aerial vehicle inspection image intelligent analysis platform

Abstract

The invention relates to the technical field of photovoltaic construction detection, and discloses a photovoltaic construction unmanned aerial vehicle inspection image intelligent analysis platform. The platform collects real-time image, position and environment parameter information of a construction target through an unmanned aerial vehicle; a historical inspection image data set is retrieved, image features are classified, historical defects and analysis data are extracted, and a historical analysis score set is obtained and optimized into a historical analysis score sequence; calculating a change rate and a stability parameter based on the sequence analysis trend; and correcting the parameters by combining the real-time environmental parameters and the standard environmental parameter deviation value of the matched feature category, and finally generating and outputting an analysis report scheme. The platform combines historical data and real-time environmental parameters for analysis and decision, improves the accuracy and effectiveness of photovoltaic construction inspection analysis, and is suitable for quality detection and analysis in the photovoltaic construction process.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic construction detection, and in particular to an intelligent analysis platform for photovoltaic construction drone inspection images. Background Art

[0002] With the rapid development of the photovoltaic industry, the scale of photovoltaic construction projects continues to expand. Quality control and defect detection during construction have become crucial for ensuring the subsequent stable operation of photovoltaic systems. Currently, inspections in photovoltaic construction areas rely heavily on manual inspections or traditional drone inspections, which have numerous limitations in practical application. Manual inspections are not only labor-intensive and time-consuming, but also significantly impacted by the inspector's experience, skill level, and subjective judgment. This can easily lead to missed inspections and false positives, making it difficult to ensure the accuracy and consistency of inspection results. Furthermore, manual inspections struggle to provide comprehensive and efficient coverage of the construction area, especially for photovoltaic construction sites with complex terrain or vast areas, where low inspection efficiency is particularly prominent. While traditional drone inspections have improved inspection coverage and efficiency to a certain extent, they suffer from significant deficiencies in image analysis and processing. Existing image analysis methods often rely on manual post-processing of drone-collected images, lacking intelligent analytical tools and unable to quickly and accurately identify defects in construction targets. Furthermore, traditional methods often focus solely on real-time image data, ignoring the value of historical inspection data. This inability to integrate historical data for trend analysis makes it difficult to predict potential problems during construction, hindering the timely implementation of effective countermeasures. The photovoltaic construction environment is complex and ever-changing. Different environmental parameters (such as light, temperature, and humidity) can significantly impact construction quality and defect detection. Existing technologies fail to fully consider the correlation between real-time environmental parameters and historical data during analysis, resulting in reduced accuracy and reliability of analysis results, making it difficult to meet the high-quality and efficient inspection requirements of photovoltaic construction. Summary of the Invention

[0003] The purpose of the present invention is to provide an intelligent analysis platform for photovoltaic construction drone inspection images to solve the problems raised in the above background technology.

[0004] To achieve the above objectives, the present invention provides an intelligent analysis platform for photovoltaic construction drone inspection images, the platform comprising: An image acquisition module is used to collect real-time image data of construction targets in the photovoltaic construction area through a drone, and obtain real-time location information and real-time environmental parameter information of the construction targets; A historical data indexing module is used to retrieve a historical inspection image data set within a user's historical time period based on the identification information of the construction target; a feature classification module for classifying the image feature information in the historical inspection image data set to generate a plurality of classification feature categories, extracting a historical defect information set and a historical analysis data set under the plurality of classification feature categories, and processing the obtained historical analysis score sets; an analysis score optimization processing module, configured to optimize the plurality of historical analysis score sets respectively to obtain a plurality of optimized analysis score sets, and arrange them in time series to obtain a plurality of historical analysis score sequences; a trend analysis module, configured to perform a trend analysis operation based on the plurality of historical analysis score sequences and calculate a rate of change parameter and a stability parameter; a correction module, configured to perform a matching operation on the real-time environmental parameter information and the plurality of classified feature categories to obtain a matching feature category, and to perform correction calculations on the change rate parameter and the stability parameter based on a deviation between the real-time environmental parameter information and the standard environmental parameter information of the matching feature category to obtain a corrected change rate parameter and a corrected stability parameter; The decision output module is used to perform analysis and decision operations based on the modified change rate parameter and the modified stability parameter, generate an analysis report solution, and output an analysis report.

[0005] Preferably, the image acquisition module collects real-time image data of the construction target in the photovoltaic construction area through a drone, and obtains real-time location information and real-time environmental parameter information of the construction target, including: The real-time image data of the construction target is captured by the image sensor of the drone, and the real-time location information of the construction target is recorded by the positioning equipment of the drone. The real-time environmental parameter information of the construction target is collected by the integrated environmental sensor, where the real-time environmental parameter information includes light intensity data, temperature data and humidity data.

[0006] Preferably, the feature classification module classifies the image feature information in the historical inspection image data set to generate multiple classification feature categories, and extracts the historical defect information set and the historical analysis data set under the multiple classification feature categories to obtain multiple historical analysis score sets, including: Extracting multiple image feature information from the historical inspection image data set, performing clustering and classification operations, and generating multiple classification feature categories; Extracting a set of historical defect information under the multiple classification feature categories, and obtaining a set of historical detection result information and a set of historical submission time submitted by users under the multiple classification feature categories; Classify and generate a plurality of basic analysis score sets according to the degree of difference between the historical detection result information set and the historical defect information set; An adjustment calculation is performed on the multiple basic analysis score sets using a proportional relationship between a preset time threshold parameter and the historical submission time set to generate multiple historical analysis score sets.

[0007] Preferably, the analysis score optimization processing module optimizes the multiple historical analysis score sets respectively to obtain multiple optimized analysis score sets, and arranges them in time series to obtain multiple historical analysis score sequences, including: selecting one of the plurality of historical analysis score sets as a current analysis score set, and determining a benchmark analysis score in the current analysis score set; Assigning weight coefficients based on the distance between other analysis scores in the current analysis score set and the benchmark analysis score to generate a basic weight distribution, wherein the distance is inversely proportional to the weight coefficient; Based on the basic weight distribution, performing an optimization compression operation on the current analysis score set to obtain an optimized analysis score set; Sorting the timestamp information of the multiple analysis scores in the optimized analysis score set to generate a historical analysis score sequence; The optimized compression operation and the sorting operation are repeatedly performed on other historical analysis score sets to generate multiple historical analysis score sequences.

[0008] Preferably, the analysis score optimization processing module performs an optimization compression operation on the current analysis score set based on the basic weight distribution to obtain an optimized analysis score set, including: Randomly selecting a preset number of analysis scores from the current analysis score set to form an initial optimized score set; Assigning weight coefficients based on the distance between the analysis scores in the initial optimization score set and the benchmark analysis scores to generate an initial weight distribution; Calculating a similarity measure between the initial weight distribution and the basic weight distribution as an optimization fitness value; Again randomly selecting a preset number of analysis scores from the current analysis score set to form a candidate optimization score set, and calculating corresponding candidate fitness values; The optimization compression operation is continuously performed until the fitness value converges, and the candidate optimization score set with the largest optimization fitness value is output as the optimization analysis score set.

[0009] Preferably, the trend analysis module performs trend analysis operations based on the multiple historical analysis score sequences to calculate the rate of change parameter and the stability parameter, including: Collect sample inspection data sets of multiple users, obtain sample analysis score sequence sets, and calculate sample change rate sets and sample stability sets based on the analysis score change characteristics of each sample analysis score sequence; Using the sample analysis score sequence set as input data, and the sample change rate set and the sample stability set as output data, constructing a trend analysis model; Applying the trend analysis model to process the plurality of historical analysis score sequences to obtain a plurality of characteristic change rate parameters and a plurality of characteristic stability parameters; Analyze the similarity values ​​between the real-time environmental parameter information and the multiple classified feature categories, perform weighted sum calculation on the multiple feature change rate parameters and the multiple feature stability parameters according to the sizes of the multiple similarity values, and generate change rate parameters and stability parameters.

[0010] Preferably, the correction module matches the real-time environmental parameter information with the multiple classification feature categories to obtain a matching feature category, and performs correction calculation on the change rate parameter and the stability parameter based on the deviation between the real-time environmental parameter information and the standard environmental parameter information of the matching feature category to obtain a corrected change rate parameter and a corrected stability parameter, including: Selecting the classification feature category with the largest similarity value as the matching feature category, and obtaining standard environmental parameter information of the matching feature category; Setting an adjustment coefficient according to a deviation between the real-time environmental parameter information and the standard environmental parameter information of the matching feature category; The adjustment coefficient is applied to perform correction calculation on the change rate parameter and the stability parameter to generate a corrected change rate parameter and a corrected stability parameter.

[0011] Preferably, the decision output module performs analysis and decision-making operations based on the modified change rate parameter and the modified stability parameter, generates an analysis report solution, and outputs an analysis report, including: Obtaining a set of sample correction rate of change parameters and a set of sample correction stability parameters, and setting a sample reporting scheme based on the magnitude of each sample correction rate of change parameter and sample correction stability parameter to generate a set of sample reporting schemes, wherein the sample reporting scheme includes a prompt level parameter, and the magnitudes of the sample correction rate of change parameter and the sample correction stability parameter are inversely proportional to the magnitude of the prompt level parameter; Using the sample correction rate of change parameter set and the sample correction stability parameter set as decision input data and the sample report scheme set as decision output data, a decision model is constructed; The decision model is applied to process the modified change rate parameter and the modified stability parameter to generate an analysis report solution.

[0012] Preferably, the platform further comprises a visualization module, the visualization module being configured to receive the analysis report solution generated by the decision output module, convert the analysis report solution into visual chart information, and output the information to a user interface; The visualization module is connected to the decision output module, and the analysis report scheme output by the decision output module serves as input data of the visualization module. The visualization module processes the analysis report scheme to generate visualization chart information.

[0013] Preferably, the platform further comprises: An environmental interference filtering module is configured to receive real-time environmental parameter information collected by the image acquisition module, identify interference scenarios such as sudden changes in light intensity, abnormal dust concentration, or sudden increases in humidity, and generate environmental interference tag data; an analysis and compensation module, configured to locate the contaminated image area according to the environmental interference mark data, calculate a compensation coefficient using the corrected stability parameter output by the correction module, and perform noise reduction and enhancement processing on the image features of the contaminated area; A dynamic weight allocation module is used to compare the image feature data after noise reduction processing with the historical defect information set generated by the feature classification module, dynamically allocate analysis weights for different defect types according to the construction stage, and output weighted defect analysis results to the decision output module.

[0014] Compared with the prior art, the present invention has the following beneficial effects: By using drones to capture real-time image data, location information, and environmental parameters, the system dynamically captures the photovoltaic construction area, providing timely visibility into the real-time status of construction targets. The historical data indexing module retrieves historical inspection image data, and combined with the feature classification module to analyze image features, it systematically integrates past defect information and analysis data to form a coherent set of historical analysis scores.

[0015] The analysis score optimization module optimizes the score set and arranges it in time series, allowing historical data to clearly display a pattern of change. This provides consistent data support for the trend analysis module's calculation of rate of change and stability parameters, facilitating an intuitive understanding of the long-term evolution of construction objectives. The correction module incorporates real-time environmental parameters. By matching feature categories and adjusting parameters based on deviations, it eliminates the impact of environmental differences on analysis results, ensuring that parameters are more aligned with actual construction scenarios. The decision-making output module generates analysis reports based on the revised parameters, transforming complex data processing results into directly applicable reference content, enabling accurate identification of potential problems and development trends during construction. The entire process, from data collection to analysis and decision-making, forms a closed loop, organically integrating real-time information with historical data, taking into account both current conditions and past trends. This makes the analysis process more coherent and comprehensive, and the output reports more closely aligned with the actual needs of photovoltaic construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a working principle diagram of the photovoltaic construction UAV inspection image intelligent analysis platform described in the present invention; Figure 2 Flowchart of the feature classification module; Figure 3 Flowchart of the processing module for analyzing score optimization; Figure 4 This is the flow chart of the trend analysis module; Figure 5 Flowchart of the correction module. DETAILED DESCRIPTION

[0017] 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.

[0018] See also Figure 1 The present invention provides a photovoltaic construction UAV inspection image intelligent analysis platform, the platform comprising: The image acquisition module is used to collect real-time image data of construction targets in the photovoltaic construction area through the UAV, and obtain the real-time location information and real-time environmental parameter information of the construction targets.

[0019] The historical data index module is used to retrieve a historical inspection image data set within a user's historical time period according to the identification information of the construction target.

[0020] The feature classification module is used to classify the image feature information in the historical inspection image data set, generate multiple classification feature categories, and extract the historical defect information set and historical analysis data set under the multiple classification feature categories to obtain multiple historical analysis score sets.

[0021] The analysis score optimization processing module is used to optimize the multiple historical analysis score sets respectively to obtain multiple optimized analysis score sets, and arrange them in time series to obtain multiple historical analysis score sequences.

[0022] The trend analysis module is used to perform a trend analysis operation based on the multiple historical analysis score sequences to calculate a change rate parameter and a stability parameter.

[0023] The correction module is used to match the real-time environmental parameter information with the multiple classification feature categories to obtain the matching feature category, and to perform correction calculations on the change rate parameter and the stability parameter based on the deviation between the real-time environmental parameter information and the standard environmental parameter information of the matching feature category to obtain the corrected change rate parameter and the corrected stability parameter.

[0024] The decision output module is used to perform analysis and decision operations based on the modified change rate parameter and the modified stability parameter, generate an analysis report solution, and output an analysis report.

[0025] The above modules are connected through a data bus to ensure data circulation and processing continuity; the input source of image data is the drone sensor array, and the output report is transmitted to the user terminal through the communication interface.

[0026] Example 1: See Figure 2 The image acquisition module captures real-time image data of construction targets using a multispectral imaging device onboard the drone. This imaging device, which includes visible and infrared channels, captures full-view images of the PV panel installation status, cable connection points, and support structures at a rate of 25 frames per second. The positioning device integrates RTK-GPS and an inertial measurement unit to record the real-time location of the construction target. Location coordinates include three-dimensional geographic coordinates (longitude, latitude, and elevation), attitude angles (pitch, yaw, and roll), and timestamp data. The environmental sensor suite, consisting of a photometer, temperature and humidity probe, and barometer, collects real-time environmental parameter information in a standardized format: light intensity is continuously recorded in lumens per square meter, temperature data is accurate to ±0.5°C, and humidity data is expressed as percentage relative humidity, correlated with the acquisition time. All sensor data is transmitted via the drone's CAN bus to an onboard processor. The processor executes a data time synchronization algorithm to align image frames, location coordinates, and environmental parameter time stamps. The data is then packaged into structured data frames for transmission to the platform server via a 4G / 5G link. The hardware layer uses an anti-shake gimbal to stabilize the image sensor to reduce image blur caused by flight vibration; the environmental sensor group is equipped with a windproof and dustproof cover to reduce the interference of external factors on sampling accuracy.

[0027] After receiving the historical inspection image data set, the feature classification module first performs multi-scale feature extraction: image keypoints are detected using the SIFT operator, texture information is quantified using the SURF descriptor, and edge gradient distribution is extracted using the HOG descriptor. For photovoltaic panel crack feature extraction, an adaptive threshold segmentation algorithm is used to distinguish crack pixel regions and calculate crack length, number of branches, and strike angle. Spot feature extraction uses HSV color space analysis to identify abnormal areas exceeding a set color gamut threshold and calculate their area percentage. Connector looseness features are identified by template matching, and the offset distance between the reference template and the real-time image is compared to output a looseness index. Subsequently, a DBSCAN clustering-based classification operation is performed, with a neighborhood radius parameter of 5 pixels and a minimum sample size dynamically adjusted to 0.1% of the total data volume. Three feature categories are generated: photovoltaic panel surface defects, bracket structural deformation, and electrical interface anomalies. Each category is associated with a historical defect information set: the crack feature category stores the mean, maximum, and growth trend of crack lengths detected over the past six months; the spot category archives a historical matrix of the dirt area percentage for each region; and the connector category retains a time-series record of torque offset distance.

[0028] The user interaction database retrieves historical inspection result information for three feature categories, including the coordinates of user-marked defect locations, severity levels (categorized as minor, moderate, and severe), and maintenance action records. A historical submission time set is extracted from the operation log, containing the start and completion timestamps of each inspection. Difference calculations utilize a two-way comparison mechanism: the IoU overlap between the user-marked crack locations and the algorithm-identified crack areas is calculated; the user-rated stain severity levels are mapped and calibrated against the algorithm-calculated stain area scores. When outputting the basic analysis score set, the crack category score is set based on the overlap deviation (overlap > 90% = 100 points, 70-90% = 80 points, and <70% = 60 points). The stain category score is calculated based on the correlation coefficient between the area percentage and the user rating (a coefficient > 0.9 = 90 points, 0.7-0.9 = 75 points). The connector category score is calculated based on the reverse conversion of the mean squared error between the offset distance and the user's judgment result. The preset time threshold parameter is a 15-day period. The adjustment calculation is performed in two steps: first, a time decay factor is established. The difference between the submission time and the current time is set as ΔT, and the decay factor η is calculated as 1-ln(ΔT+1) / ln(16). Then, the factor is applied to adjust the basic score. If ΔT≤15 days, the adjusted score = basic score × (1+η / 2); if ΔT>15 days, the score = basic score × (1-η / 2). The final three types of historical analysis score sets are stored in the distributed database according to the feature category number.

[0029] In terms of data stream processing, real-time images from the image acquisition module are compressed via edge computing nodes, using H.265 encoding to reduce bandwidth usage. The clustering and classification operations of the feature classification module are deployed on a GPU cluster, using CUDA parallelism to accelerate feature vector distance calculations. An anomaly detection mechanism is incorporated into the generation of historical analysis score sets: when the user-submitted timestamp deviates from the sensor recording time by more than 300 seconds, the data verification process is triggered to recalibrate the time series. In environmental parameter information processing, temperature data uses a sliding average filter to eliminate the impact of transient fluctuations, and light intensity data is fused with redundant sampling values ​​from multiple sensors to improve reliability. Inter-module data exchange utilizes the Apache Avro serialization protocol, with image data transmitted as a binary stream, feature vectors encapsulated in Float array format, and analysis score sets stored in a JSON structure. The platform service layer establishes an independent data sandbox for each construction target to prevent data cross-interference during multi-tasking. The hardware infrastructure utilizes containerized deployment, with the feature classification module running within a Docker container, using shared memory to accelerate image feature transmission.

[0030] Example 2: See Figure 3 When the analysis score optimization module starts, it loads multiple historical analysis score sets generated by the feature classification module from the distributed database. Each set corresponds to defect analysis records for a specific construction phase or component type. During the module initialization phase, a data selection strategy is set: the median distribution of all score sets is traversed, and the set with the narrowest fluctuation range is selected as the current analysis score set. The baseline analysis score calculation uses robust processing logic, first removing the top and bottom 10% of extreme values ​​in the current set. The arithmetic mean of the remaining data is then set as the baseline value.

[0031] The weight assignment process performs a discrete degree mapping transformation, calculating the absolute difference between each analysis score in the current set and the baseline value. A difference quantization table is pre-set in the module memory: if the absolute difference is less than 5% of the baseline value, the weight coefficient is set to 1.0; differences between 5% and 10% of the baseline are assigned a coefficient of 0.8; differences between 10% and 15% are assigned a coefficient of 0.6; differences between 15% and 20% are assigned a coefficient of 0.4; and differences exceeding 20% ​​have their weight reduced to 0.1. The weight vector is constructed using a linearly decreasing principle. All coefficients are normalized to form a basic weight distribution matrix, whose dimensions correspond to the capacity of the current analysis score set.

[0032] The optimization compression operation initiates a Monte Carlo iteration process, initially with a pre-set selection ratio of one-fifth of the total sample set. A random number generator generates an initial index sequence, extracting the analysis scores at the corresponding positions to form an initial set of optimized scores. Each score in this set is recalculated by comparing its absolute difference with the baseline value, and the same rules are used to generate an initial weight distribution. A similarity metric uses the vector angle principle, converting the base weight distribution and the initial weight distribution into unit vectors. The dot product result is defined as the optimized fitness value. This value, rounded to four decimal places, is stored in the iteration buffer.

[0033] Multiple rounds of optimization screening are performed continuously, generating a new random index sequence each time to construct a candidate optimization score set. After the weight distribution of the candidate set is reconstructed, vector similarity comparisons are repeated with the base weight distribution. Parallel processing units monitor the trend of the optimization fitness values, and convergence is determined when the change in fitness values ​​between two consecutive optimizations is less than one thousandth. Finally, the candidate set with the highest fitness value during the survival period is selected as the optimization analysis score set. The elimination process cumulatively compares the results of eight candidate sets.

[0034] The time series generation phase utilizes a timestamp processing engine to parse the collection time tags associated with each score from the optimized analysis score set. Timestamps are stored in UNIX millisecond timestamp format, and a converter converts them to the standard ISO8601 date format. The sorting algorithm uses monotonically increasing logic for all dates, sorting data from the same year, month, and day in chronological order with millisecond precision. The generated historical analysis score sequence is stored in a linked list structure, with nodes containing a date and time string and the corresponding analysis score value. The module incorporates an internal self-checking mechanism to verify the continuity and logical order of timestamps within the sequence, triggering a reordering process if a time reversal is detected.

[0035] The module's hardware implementation is based on a heterogeneous computing architecture: the control unit uses a dual-core ARM processor to schedule the overall process, weight calculations and vector operations are deployed on the NVIDIA Jetson GPU unit, and the iterative optimization and compression process is accelerated by FPGA logic circuits. The data channel consists of two types of buses: a control bus that transmits module instruction status, and a high-speed data bus that transmits floating-point matrices of historical analysis score sets. The processing cycle sets a threshold, and the optimization and compression of a single set is completed within 300 milliseconds. The storage interface is compatible with SATA solid-state drives and is used to cache intermediate data during the iterative process. The exception handling unit monitors the operating status and automatically switches to redundant data processing mode to continue execution when it detects anomalies in the score set data or missing timestamps.

[0036] The output layer encapsulates the historical analysis score sequence into a specific data structure, with metadata describing each sequence: the corresponding classification feature category number, the construction target area code, and the sequence generation time. Data packets are transmitted to the trend analysis module's input queue via a fiber optic channel, using an industrial-grade Time-Sensitive Network (TSN) protocol to ensure timing accuracy. A resource recycling mechanism automatically releases temporary storage space after a sequence is successfully transmitted, freeing up computing resources for processing the next score set. Module initialization parameters support dynamic configuration, allowing the baseline value calculation rules and weight distribution gradients to be adjusted based on different PV scenarios.

[0037] Example 3: See Figure 4 and Figure 5 The operation of the trend analysis module begins with the sample data preprocessing stage, loading a sample inspection data set of multiple users from the distributed storage system. Each sample contains analysis score records of no less than 30 consecutive time points. In the data cleaning stage, abnormal sample points are eliminated, and the sliding window method is used to detect data points that deviate from the mean by more than three times the standard deviation, and then replace them with linear interpolation of adjacent time points. The sample analysis score sequence set is grouped by photovoltaic module type. Each sequence is normalized and the original score is mapped to the [0,1] interval. The change rate feature extraction adopts the piecewise linear fitting method. Each sample sequence is divided into five equal-length time periods, and the slope of each segment is calculated as the local change rate. The stability feature is characterized by the sliding coefficient of variation. The window width is set to 20% of the total sequence length, and the ratio of the standard deviation of the data in the window to the mean is calculated.

[0038] The trend analysis model is constructed using a two-layer LSTM neural network architecture. The input layer accepts fractional sequence segments of length 10, the hidden layer has 64 memory units, and the output layer contains two independent fully connected branches, one for predicting the rate of change parameter and the other for predicting the stability parameter. The model is trained using an adaptive moment estimation optimizer, with an initial learning rate of 0.001 and a batch size of 32 samples. To prevent overfitting, a random dropout layer with a dropout rate of 0.2 is added after the hidden layer. A composite loss function is designed, using the Huber loss for the rate of change branch and the log-cosine loss for the stability branch. Training is continuously monitored using the early stopping metric on the validation set, and training is terminated when the validation loss fails to decrease for five consecutive epochs.

[0039] During the real-time analysis phase, the module receives a sequence of historical analysis scores from the analysis score optimization processing module. Each sequence is first normalized in the same way as the training data. The model inference engine loads the pretrained weight file and performs forward propagation calculations on the input sequence. The rate of change parameter of the output layer is converted to its original dimension and multiplied by 100 to express it as a percentage change rate. The stability parameter is maintained as a dimensionless value in the interval [0, 1]. During the similarity calculation phase, the matching of real-time environmental parameter information with the classification feature category is performed using a multidimensional spatial distance metric:

[0040] in, Represents the similarity value, Represents the real-time environment parameters dimensions (light intensity, temperature, humidity), The first standard environmental parameter for classifying characteristic categories dimensions, is the weight coefficient for each dimension (0.5 for light, 0.3 for temperature, and 0.2 for humidity). The calculation results in a similarity vector with the same number of dimensions as the classification feature categories.

[0041] During the parameter fusion phase, weighted aggregation is performed on multiple feature rate of change and stability parameters output by the model. Weights are assigned based on the ranking of similarity values: the parameter corresponding to the category with the highest similarity is assigned a weight of 0.6, the parameter corresponding to the category with the second highest similarity is assigned a weight of 0.3, and the remaining categories share the remaining weight of 0.1. The weighted summation of the rate of change and stability parameters is then range-calibrated. The rate of change parameter is limited to the interval [-50%, 50%], with the boundary value being used when exceeding the threshold. The stability parameter is adjusted to the range of 0.2-0.8 using a sigmoid function.

[0042] The correction module's operational process identifies the index corresponding to the maximum value in the similarity vector and sets the classification feature category pointed to by this index as the matching feature category. Standard environmental parameter information is extracted from the feature category metadata, including baseline values ​​for light intensity (unit: klx), temperature (unit: °C), and humidity (unit: %RH). Deviations are calculated using the relative difference method. The absolute difference between the real-time value and the baseline value for each of the three environmental parameters is calculated and then divided by the baseline value to obtain the normalized deviation. The adjustment coefficient is generated using a piecewise function strategy: when the total deviation is less than 0.1, the coefficient is 1; when the deviation is between 0.1 and 0.3, the coefficient is 0.9; when the deviation is between 0.3 and 0.5, the coefficient is 0.7; and when the deviation exceeds 0.5, the coefficient is reduced to 0.5.

[0043] The correction calculation process performs differentiated processing on the rate-of-change parameter and the stability parameter. The rate-of-change parameter is corrected using a multiplication model: the original parameter is multiplied by the adjustment coefficient, and then a linear compensation term for the deviation is added. The stability parameter is corrected using a nonlinear transformation, using the adjustment coefficient as the base of the exponential term and the original parameter as a power. Corrected parameter values ​​retain three decimal places of precision, and the data validation module verifies the numerical rationality. Values ​​outside the theoretical range trigger a recalculation process.

[0044] In terms of hardware implementation, the module is deployed on a dedicated inference server equipped with an NVIDIA T4 GPU to accelerate model operations. The environmental parameter processor utilizes a low-power ARM architecture to independently process sensor data streams. A double-buffered memory mechanism ensures a continuous data supply during real-time analysis. An exception handling unit monitors the entire process and automatically switches to a backup analysis mode when it detects sensor data interruptions or model inference timeouts. The communication interface utilizes a time-triggered Ethernet protocol to ensure the timing determinism of correction parameter transmission.

[0045] The data persistence layer records complete correction logs, including original parameter values, environmental deviations, adjustment coefficients, and correction results. Log entries are timestamped and identified by the operator, supporting subsequent audit trails. Visualizing intermediate data allows for detailed similarity calculations and weight assignments in debug mode, assisting with algorithm optimization and verification. Module configuration parameters support hot updates, allowing for dynamic adjustment of deviation thresholds and correction strategies through the management interface.

[0046] The output interface encapsulates the modified rate of change parameters and the modified stability parameters into structured messages, appending data quality indicators and confidence scores. The message queue producer pushes the data to the subscription topic of the decision output module and simultaneously writes it to the time series database for long-term trend analysis. After each analysis, the resource recycling thread cleans up temporary data, resets the model inference state, and prepares for the next processing task.

[0047] Example 4: The decision output module begins its operation by receiving parameters. The modified rate of change parameter records the trend of construction defects as a percentage (positive values ​​indicate deterioration, negative values ​​indicate improvement). The modified stability parameter describes the degree of fluctuation as a value between 0 and 1 (larger values ​​indicate greater stability). The module's built-in decision model is generated through training using a support vector machine algorithm. The training dataset contains modified parameter combinations for various scenarios and manually annotated corresponding decision solutions. The following table shows the mapping between five typical input parameter sets and output solutions: Table 1: Mapping relationship between five typical input parameters and output solutions

[0048] During the model training phase, 200 sets of sample parameters from historical maintenance records were collected as input feature vectors, and the output labels corresponded to five reporting scenarios. The support vector machine kernel used a radial basis function with a penalty coefficient set to 1.0, and the decision boundary was optimized using 5-fold cross-validation. During the real-time decision-making process, the input vector [corrected rate of change parameter, corrected stability parameter] was preprocessed with feature scaling before being fed into the model. The output layer probability distribution triggered the corresponding scenario selection. When the difference between the probabilities of two categories was less than 0.2, the multi-scenario fusion mechanism was activated to generate a combined report.

[0049] After receiving the report plan output by the decision, the visualization module executes the data-to-graphic conversion logic. For the red warning reports in the table, the module generates a heat map superimposed on the three-dimensional model of the construction area: the crack extension area is marked with a flashing red dot, and the sidebar displays a line chart for historical data comparison. The regular report is converted into an interactive dashboard. The user clicks on the cable trough number to expand the loosening trend for twelve months. The email text report is embedded with a dynamic bar chart to show the difference between the rust area of ​​the bracket and the standard threshold. The three-dimensional positioning map uses layered rendering technology, the electrical short circuit point is displayed with a pulsed light effect, and the insulation resistance parameters are displayed in the list of related equipment in a floating manner. The visualization dashboard is integrated into the web interface, and multi-level drill views are divided according to the construction area. The areas where the stain cleaning meets the standards are displayed with a green check mark.

[0050] The hardware support environment consists of a dual-core server cluster. The decision model runs on compute nodes equipped with 128GB of memory, and visualization rendering tasks are distributed to four GPU accelerator cards. When processing the 3D positioning map, the OpenGL pipeline generates a photovoltaic station model containing more than 20,000 triangles in real time, and the texture maps are loaded with drone orthophotos. The network transport layer uses a layered compression strategy: structured report data is transmitted in Protobuf format, averaging 8KB of bytes; the heat map data stream is compressed to 30% of its original size using Delta encoding; and the 3D model uses a level-of-detail (LOD) loading mechanism, initially transmitting only a simplified topology.

[0051] The message routing system selects delivery channels based on the reporting scheme type: red alert reports trigger push notifications via SMS gateways, email servers, and mobile apps; regular reports are written to a relational database and retrieved by the platform's message center. Two-factor authentication is used for security, requiring a second biometric verification of the engineering supervisor when issuing shutdown and maintenance instructions. Data archiving is done in separate storage volumes based on PV project numbers. All historical reports are linked to snapshots of environmental parameters at the corresponding time, enabling retrospective reconstruction of decision-making scenarios.

[0052] The user interface design incorporates multi-level control features. Hovering the mouse over a stain-compliant area in the visual dashboard displays a timeline of cleaning records. A slider in the 3D model view allows for time-based playback of defect evolution. The report export menu provides downloadable evidence chains, from raw inspection data to decision-making reports. All graphical elements adhere to WCAG 2.0 accessibility standards, and color-blind mode automatically converts red and green warning signs to zebra-striped fill patterns. The real-time monitoring dashboard refresh rate is set to 5 seconds, triggering a pop-up notification when the correction rate of change parameter exceeds 15%.

[0053] A version control mechanism ensures traceability of decision rules: After each model update, the new decision logic is tested in a 10% end-user environment via a phased release. Historical decision records are stored alongside current rules, and audit mode allows comparison of decision differences between different rule versions for the same parameters. A three-level failover strategy is designed: automatic failover to a backup model service in the event of a primary server outage; in the event of a full system failure, offline decision tables are used to generate basic reports; and in the event of network unavailability, data is cached on a local SSD and retransmitted in batches upon reconnection.

[0054] Example 5: The environmental interference filtering module continuously receives the real-time environmental parameter data stream transmitted by the image acquisition module, and the input channel samples the light intensity data, temperature data and humidity data at a frequency of 100Hz. The light intensity mutation detection adopts a two-way threshold judgment mechanism: when the light change amplitude exceeds 20% of the baseline value in three consecutive sampling cycles, and does not return to the baseline range in the subsequent two cycles, the light anomaly mark is triggered; the dust concentration monitoring relies on a scattering optical sensor, and the dust mark is activated when the standard deviation of the sudden increase in the number of suspended particles is greater than three times the historical mean; the humidity sudden increase judgment is based on the humidity gradient algorithm, and the humidity anomaly is recorded when the change rate exceeds 10% per minute. The tag data structure includes the interference type code, the starting timestamp, the duration and the spatial positioning coordinates, and the coordinates are derived from the location information associated with the image acquisition module. The generated environmental interference tag data is embedded in the message header and transmitted to the analysis and compensation module.

[0055] The analysis and compensation module polls the marker message queue in real time. Upon detecting an interference marker, it immediately loads the original image corresponding to the time window. The region positioning engine delineates a rectangular detection box in the drone's orthophoto coordinate system based on the coordinate parameters in the marker. Contaminated areas are identified using adaptive mask generation technology. For scenes with sudden changes in illumination, pixel clusters exceeding ±30% of the ambient brightness mean are segmented. For dust interference, foggy areas with color saturation below 25% are extracted. For humidity interference, abnormal areas with hue shifted to the blue spectrum are located. Segmentation boundaries are stored as vector polygons, with vertex coordinates mapped to construction target entities. The compensation coefficient calculation module uses the correction stability parameters from the correction module and scales them as input variables for an exponential function. Image enhancement uses a channel-by-channel operation: histogram equalization is performed on the red channel of the contaminated area to compensate for color distortion, bilateral filtering is applied to the blue channel to suppress water mist noise, and the non-local means algorithm is used to restore detailed texture in the green channel. The processed local image undergoes edge feathering and is reconstructed into the original image using alpha blending to output the compensated image.

[0056] The dynamic weight assignment module initiates the defect comparison process, loading the compensated image feature data and the historical defect information stored by the feature classification module. The feature extractor extracts the structural crack morphology vector, the electrical component position topology, and the surface stain distribution matrix from the compensated image. The comparison unit performs a three-stage matching process: Hausdorff distance calculation between the structural crack feature and the maximum crack morphology in the historical database; offset estimation based on topological similarity for electrical component position matching; and spatial correlation coefficient calculation of stain distribution through matrix convolution. The construction stage identifier parses the drone inspection mission codes: 01-04 for the pile foundation construction phase, 05-08 for the bracket installation phase, and 09-12 for the component commissioning phase. The default defect weighting strategy for each phase is 0.8 for structural deformation in the pile foundation phase and 0.7 for electrical defects in the component phase. The dynamic assignment algorithm reads the phase code matching strategy table, loads baseline weights based on defect type, and then superimposes the influence coefficients on the comparison results. Crack defect weights are adjusted based on the Hausdorff distance reduction ratio, with a 0.15 weight increase when the distance is less than 20% of the historical value. Electrical defect weights are gradient-corrected based on the topological offset. Stain weights are multiplied by the square root of the correlation coefficient. The weighted defect analysis results are packaged as a structured dataset with fields containing the defect type code, original score, dynamic weight value, and final weighted score for each construction point. The data packets are time-stamped and transmitted to the preparatory analysis interface of the decision output module.

[0057] Multi-level redundant control is deployed at the system implementation level. The hot standby node of the environmental interference filtering module takes over the data stream when the host response delay exceeds 50ms. The image processing tasks of the analysis and compensation module are distributed to a heterogeneous computing cluster. The CPU handles regional segmentation tasks, the GPU cluster performs channel enhancement operations, and the FPGA accelerates hybrid recombination operations. The dynamic weight allocation module uses a distributed in-memory database to store stage policy tables, supporting thousands of concurrent read requests. Communication transmission adopts a fiber-optic ring topology, and point-to-point encrypted data channels are established between modules. The transmission protocol applies the security standards of the Industrial Internet of Things. The exception handling unit monitors the entire workflow and initiates the default compensation mode based on historical averages when it detects the loss of interference markers or compensation calculation timeouts.

[0058] The persistent storage system records the entire operation chain: from the original sampling value of environmental interference to the final weighted defect score, the complete data chain is partitioned and stored by construction area. The visual debugging interface allows operation and maintenance personnel to view the interference area segmentation effect, image comparison before and after compensation, and weight distribution calculation tree. System maintenance supports online policy updates. After the engineering construction plan is changed, the new stage weight comparison table can be imported through the configuration interface. The change record generates an independent version number and is archived in the engineering knowledge base. Execution efficiency optimization measures include a pre-processing cache mechanism, which pre-loads interference area templates that appear frequently within three months to accelerate the positioning process. The lifecycle management module automatically archives historical policy data sets that have not been accessed for more than three years, and retains policies compressed into minimum feature snapshots for audit and tracing.

[0059] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0060] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent analysis platform for photovoltaic construction drone inspection images, characterized by: The platform includes: An image acquisition module is used to collect real-time image data of construction targets in the photovoltaic construction area through a drone, and obtain real-time location information and real-time environmental parameter information of the construction targets; A historical data indexing module is used to retrieve a historical inspection image data set within a user's historical time period based on the identification information of the construction target; a feature classification module for classifying the image feature information in the historical inspection image data set to generate a plurality of classification feature categories, extracting a historical defect information set and a historical analysis data set under the plurality of classification feature categories, and processing the obtained historical analysis score sets; an analysis score optimization processing module, configured to optimize the plurality of historical analysis score sets respectively to obtain a plurality of optimized analysis score sets, and arrange them in time series to obtain a plurality of historical analysis score sequences; a trend analysis module, configured to perform a trend analysis operation based on the plurality of historical analysis score sequences and calculate a rate of change parameter and a stability parameter; a correction module, configured to perform a matching operation on the real-time environmental parameter information and the plurality of classified feature categories to obtain a matching feature category, and to perform correction calculations on the change rate parameter and the stability parameter based on a deviation between the real-time environmental parameter information and the standard environmental parameter information of the matching feature category to obtain a corrected change rate parameter and a corrected stability parameter; The decision output module is used to perform analysis and decision operations based on the modified change rate parameter and the modified stability parameter, generate an analysis report solution, and output an analysis report.

2. The photovoltaic construction drone inspection image intelligent analysis platform according to claim 1 is characterized in that: The image acquisition module collects real-time image data of construction targets in the photovoltaic construction area through the UAV, and obtains real-time location information and real-time environmental parameter information of the construction targets, including: The real-time image data of the construction target is captured by the image sensor of the drone, and the real-time location information of the construction target is recorded by the positioning equipment of the drone. The real-time environmental parameter information of the construction target is collected by the integrated environmental sensor, where the real-time environmental parameter information includes light intensity data, temperature data and humidity data.

3. The photovoltaic construction drone inspection image intelligent analysis platform according to claim 1 is characterized in that: The feature classification module classifies the image feature information in the historical inspection image data set to generate multiple classification feature categories, extracts historical defect information sets and historical analysis data sets under the multiple classification feature categories, and processes to obtain multiple historical analysis score sets, including: Extracting multiple image feature information from the historical inspection image data set, performing clustering and classification operations, and generating multiple classification feature categories; Extracting a set of historical defect information under the multiple classification feature categories, and obtaining a set of historical detection result information and a set of historical submission time submitted by users under the multiple classification feature categories; Classify and generate a plurality of basic analysis score sets according to the degree of difference between the historical detection result information set and the historical defect information set; An adjustment calculation is performed on the multiple basic analysis score sets using a proportional relationship between a preset time threshold parameter and the historical submission time set to generate multiple historical analysis score sets.

4. The photovoltaic construction UAV inspection image intelligent analysis platform according to claim 1 is characterized in that: The analysis score optimization processing module optimizes the multiple historical analysis score sets respectively to obtain multiple optimized analysis score sets, and arranges them in time series to obtain multiple historical analysis score sequences, including: selecting one of the plurality of historical analysis score sets as a current analysis score set, and determining a benchmark analysis score in the current analysis score set; Assigning weight coefficients based on the distance between other analysis scores in the current analysis score set and the benchmark analysis score to generate a basic weight distribution, wherein the distance is inversely proportional to the weight coefficient; Based on the basic weight distribution, performing an optimization compression operation on the current analysis score set to obtain an optimized analysis score set; Sorting the timestamp information of the multiple analysis scores in the optimized analysis score set to generate a historical analysis score sequence; The optimized compression operation and the sorting operation are repeatedly performed on other historical analysis score sets to generate multiple historical analysis score sequences.

5. The photovoltaic construction UAV inspection image intelligent analysis platform according to claim 4 is characterized in that: The analysis score optimization processing module performs an optimization compression operation on the current analysis score set based on the basic weight distribution to obtain an optimized analysis score set, including: Randomly selecting a preset number of analysis scores from the current analysis score set to form an initial optimized score set; Assigning weight coefficients based on the distance between the analysis scores in the initial optimization score set and the benchmark analysis scores to generate an initial weight distribution; Calculating a similarity measure between the initial weight distribution and the basic weight distribution as an optimization fitness value; Again randomly selecting a preset number of analysis scores from the current analysis score set to form a candidate optimization score set, and calculating corresponding candidate fitness values; The optimization compression operation is continuously performed until the fitness value converges, and the candidate optimization score set with the largest optimization fitness value is output as the optimization analysis score set.

6. The photovoltaic construction UAV inspection image intelligent analysis platform according to claim 1 is characterized in that: The trend analysis module performs a trend analysis operation based on the multiple historical analysis score sequences to calculate a change rate parameter and a stability parameter, including: Collect sample inspection data sets of multiple users, obtain sample analysis score sequence sets, and calculate sample change rate sets and sample stability sets based on the analysis score change characteristics of each sample analysis score sequence; Using the sample analysis score sequence set as input data, and the sample change rate set and the sample stability set as output data, constructing a trend analysis model; Applying the trend analysis model to process the plurality of historical analysis score sequences to obtain a plurality of characteristic change rate parameters and a plurality of characteristic stability parameters; Analyze the similarity values ​​between the real-time environmental parameter information and the multiple classified feature categories, perform weighted sum calculation on the multiple feature change rate parameters and the multiple feature stability parameters according to the sizes of the multiple similarity values, and generate change rate parameters and stability parameters.

7. The photovoltaic construction UAV inspection image intelligent analysis platform according to claim 6 is characterized in that: The correction module matches the real-time environmental parameter information with the multiple classification feature categories to obtain a matching feature category, and performs correction calculation on the change rate parameter and the stability parameter based on the deviation between the real-time environmental parameter information and the standard environmental parameter information of the matching feature category to obtain a corrected change rate parameter and a corrected stability parameter, including: Selecting the classification feature category with the largest similarity value as the matching feature category, and obtaining standard environmental parameter information of the matching feature category; Setting an adjustment coefficient according to a deviation between the real-time environmental parameter information and the standard environmental parameter information of the matching feature category; The adjustment coefficient is applied to perform correction calculation on the change rate parameter and the stability parameter to generate a corrected change rate parameter and a corrected stability parameter.

8. The photovoltaic construction UAV inspection image intelligent analysis platform according to claim 1 is characterized in that: The decision output module performs an analysis and decision operation based on the modified change rate parameter and the modified stability parameter, generates an analysis report solution, and outputs an analysis report, including: Obtaining a set of sample correction rate of change parameters and a set of sample correction stability parameters, and setting a sample reporting scheme based on the magnitude of each sample correction rate of change parameter and sample correction stability parameter to generate a set of sample reporting schemes, wherein the sample reporting scheme includes a prompt level parameter, and the magnitudes of the sample correction rate of change parameter and the sample correction stability parameter are inversely proportional to the magnitude of the prompt level parameter; Using the sample correction rate of change parameter set and the sample correction stability parameter set as decision input data and the sample report scheme set as decision output data, a decision model is constructed; The decision model is applied to process the modified change rate parameter and the modified stability parameter to generate an analysis report solution.

9. The photovoltaic construction UAV inspection image intelligent analysis platform according to claim 1 is characterized in that: The platform further includes a visualization module, which is used to receive the analysis report solution generated by the decision output module, convert the analysis report solution into visual chart information, and output it to the user interface; The visualization module is connected to the decision output module, and the analysis report scheme output by the decision output module serves as input data of the visualization module. The visualization module processes the analysis report scheme to generate visualization chart information.

10. The photovoltaic construction UAV inspection image intelligent analysis platform according to claim 1 is characterized in that: Also includes: An environmental interference filtering module is configured to receive real-time environmental parameter information collected by the image acquisition module, identify interference scenarios such as sudden changes in light intensity, abnormal dust concentration, or sudden increases in humidity, and generate environmental interference tag data; an analysis and compensation module, configured to locate the contaminated image area according to the environmental interference mark data, calculate a compensation coefficient using the corrected stability parameter output by the correction module, and perform noise reduction and enhancement processing on the image features of the contaminated area; A dynamic weight allocation module is used to compare the image feature data after noise reduction processing with the historical defect information set generated by the feature classification module, dynamically allocate analysis weights for different defect types according to the construction stage, and output weighted defect analysis results to the decision output module.

Images