photovoltaic panel Monitoring method

Abstract

The present invention relates to a solar panel monitoring method, and the technical problem to be solved is to provide a solar panel monitoring method in which an unmanned aerial vehicle automatically photographs a solar cell covered with dust and automatically analyzes the captured image to automatically detect a solar cell malfunctioning due to dust. The solar panel monitoring method of the present invention according to the present invention further comprises: an image information acquisition step in which an unmanned aerial vehicle takes a picture at a preset altitude to acquire image information about a solar panel; and an image information analysis step in which the image information is received as thermal image information and processed via video mosaic to analyze the surface temperature characteristics of the cell units of the solar panel.

Description

Technology Field

[0001] The present invention relates to a solar panel monitoring method, and more specifically, to a solar panel monitoring method capable of monitoring the malfunction status of the sun and the panel by analyzing the occurrence of dust on the solar panel using a thermal image. Background Technology

[0003] Dust accumulated on solar cells causes them to malfunction, and unmanned aerial vehicles (UAVs) are required to monitor such dust accumulated on solar cells.

[0004] In this case, the unmanned aerial vehicle must photograph the dust while flying at a very low altitude of about 30 to 50 meters due to the problem of image information resolution.

[0005] In addition, automatic flight of unmanned aerial vehicles can only be operated at a certain altitude where collision avoidance with adjacent flight obstacles (e.g., power lines, trees, buildings) can be guaranteed. Because of this, there is a problem in that unnecessary objects (buildings, roads, etc.) around the solar panels are captured in aerial photographs taken during automatic flight, occupying more than 99% of the image.

[0006] Therefore, the method of capturing photographic information by automatically flying unmanned aerial vehicles has the problem of poor data processing efficiency because most of the monitoring process is allocated to processing unnecessary data during data processing.

[0007] In addition, several studies have developed real-time mosaics using UAV video thermal imaging to detect defective solar panels in large-scale photovoltaic power plants, and another study has evaluated the suitability of UAV-based video thermal imaging in terms of mapping accuracy and thermal signals compared to autopilot-based imaging of solar panels.

[0008] However, since conventional research has not investigated a method of inspecting solar panels malfunctioning due to dust by photographing them with unmanned aerial vehicles, a technical solution is required to inspect solar panels that malfunction due to dust. Prior art literature

[0010] Korean Registered Patent Publication No. 10-1806217 (Published Dec. 07, 2017) The problem to be solved

[0011] The technical objective of the present invention to solve the aforementioned problem is to provide a solar panel monitoring method that enables an unmanned aerial vehicle to photograph a solar cell covered in dust using autopilot and to automatically detect a solar cell malfunctioning due to dust by automatically analyzing the captured image.

[0012] Furthermore, another technical objective of the present invention is to provide a solar panel monitoring method capable of performing inspections at the cell level of the solar panel, rather than a method that analyzes only images of the entire area where the solar panels are installed, in order to overcome the problem that conventional solar cell thermal defect inspection methods cannot perform inspections at the solar panel level due to a lack of key points and unnecessary data duplication.

[0013] In addition, another technical objective of the present invention is to provide a solar panel monitoring method capable of detecting spatial patterns and distribution characteristics at the level of a photo-mosaic using still images, even when using video frames with a resolution lower than that of still images when detecting spatial patterns and distribution characteristics.

[0014] In addition, another technical objective of the present invention is to provide a solar panel monitoring method that can analyze more effectively than conventional methods by using video mosaics, which can find 2.5 to 3.4 times more matching key points per solar module than photo mosaics. means of solving the problem

[0016] A solar panel monitoring method for achieving the above-mentioned technical task further comprises: an image information acquisition step in which an unmanned aerial vehicle takes a picture at a preset altitude to acquire image information about a solar panel; and an image information analysis step in which the image information acquired in the image information acquisition step is received as thermal image information and analyzed for surface temperature characteristics of the cell units of the solar panel by video-mosaic processing.

[0017] In this case, the image information analysis step further includes detecting a spatial pattern during the video-mosaic processing.

[0018] In addition, the image information acquisition step further includes acquiring the image information as video frames that are continuously captured.

[0019] In addition, in the image information analysis step, the ground sampling distance (GSD) of the image captured by the unmanned aerial vehicle is given by Equation (1). It is determined as, where Sw is the width of the camera sensor, H is the flight height, and F R ... is formed by the camera's focal length, and imW is formed by the image width; further includes.

[0020] In addition, the image information analysis step further includes geometrically correcting the captured image based on the SfM (Structure-from-Motion) algorithm.

[0021] In this case, when using the above SfM (Structure-from-Motion) algorithm, it further includes reconstructing geometric data into 3D scene shapes and camera motions using a 2D image sequence captured by a camera of an unmanned aerial vehicle.

[0022] Meanwhile, the image information analysis step further includes generating spatial information about time using an SRT file that provides total direction data in seconds, along with numbers representing sequences, Coordinated Universal Time (UTC), and total direction parameters obtained from the flight controller of the unmanned aerial vehicle.

[0023] In addition, the image information analysis step further includes: an initial processing step comprising a keypoint extraction process, a keypoint matching process, a camera model optimization process, and a GPS / GCP acquisition process for each of the images within a video frame acquired through the unmanned aerial vehicle; a point cloud and mesh generation step that converts the points of the images within the video frame acquired at the shooting point after the initial processing step into a 3D texture mesh in a high-density state; and a digital surface model (DSM) development step that indexes each of the images determined by the point cloud and mesh generation step and then performs video-mosaic processing. Effects of the invention

[0025] The present invention has the effect of automatically detecting solar cells that are malfunctioning due to dust by automatically analyzing the captured images and having an unmanned aerial vehicle photograph a solar cell covered in dust using autopilot.

[0026] In addition, the present invention has the effect of enabling more sophisticated analysis than conventional methods by allowing inspection at the cell level of solar panels, rather than analyzing only images of the entire area where solar panels are installed, thereby overcoming the problem in conventional solar cell thermal defect inspection methods where inspection cannot be performed at the cell level due to a lack of key points and unnecessary data duplication.

[0027] In addition, the present invention has the effect of detecting spatial patterns and distribution characteristics at the level of a photo-mosaic using still images, even when using video frames with a resolution lower than that of still images when detecting spatial patterns and distribution characteristics.

[0028] In addition, since the present invention uses video mosaic, it can find 2.5 to 3.4 times more matching key points per solar module than photo mosaic, thus having the effect of performing superior analysis compared to conventional methods. Brief explanation of the drawing

[0030] FIG. 1 is a flowchart of a solar panel monitoring method according to an embodiment of the present invention. FIG. 2 is a photograph of a solar panel taken using a solar panel monitoring method according to an embodiment of the present invention. FIG. 3(a) is a photo-mosaic with multiple mosaic images superimposed in a thermal mosaic of an experimental site with a point cloud, and FIG. 3(b) is a video-mosaic with multiple mosaic images superimposed in a thermal mosaic of an experimental site with a point cloud. Figure 4(a) is the differentiated distribution density of matched keypoints for a photo-mosaic of a solar cell, and Figure 4(b) is the differentiated distribution density of matched keypoints for a photo-mosaic of a solar cell. FIG. 5(a1) is an overlaid on-field photograph of a hotspot boundary (derived from a photo-mosaic) of a solar cell malfunctioning due to dust, FIG. 5(b1) is an overlaid on-field photograph of a hotspot boundary (derived from a video-mosaic) of a solar cell malfunctioning due to dust, FIG. 5(a2) is a photograph of the hotspot boundary detected in the photo-mosaic for FIG. 5(a1), and FIG. 5(b2) is a photograph of the hotspot boundary detected in the video-mosaic for FIG. 5(b1). FIG. 6(a1) is an enlarged image of a portion of the scene photograph detected in the photo-mosaic of FIG. 5(a1), FIG. 6(b1) is an enlarged image of a portion of the scene photograph detected in the video-mosaic of FIG. 5(b1), FIG. 6(a2) is a hotspot boundary enlarged of a portion of FIG. 6(a1), and FIG. 6(b2) is a hotspot boundary enlarged of a portion of FIG. 6(b1). Figure 7 shows the results of a regression analysis to investigate the solar cell surface temperature that detected a common hotspot in the video mosaic and photo mosaic of the testbed. FIG. 8(a) is a comparison graph of the solar cell surface temperature adjustment values ​​of the photo-mosaic (yellow bars) and the solar cell surface temperature adjustment values ​​of the video mosaic (blue bars) for test bed 1, and FIG. 8(b) is a comparison graph of the solar cell surface temperature adjustment values ​​of the photo-mosaic (yellow bars) and the solar cell surface temperature adjustment values ​​of the video mosaic (blue bars) for test beds 2 and 3. Specific details for implementing the invention

[0031] Hereinafter, embodiments for implementing the present invention will be described with reference to the attached drawings. In this case, whenever a part is described as "comprising" a certain component throughout the specification, unless specifically stated otherwise, this is deemed to mean that it does not control other components but may include additional components. Furthermore, terms such as "...part" described in the specification are deemed to mean a unit that processes at least one function or operation when describing electronic hardware or electronic software, and to mean a single part, function, use, point, or driving element when describing mechanical devices. Additionally, in the following description, identical or similar configurations will be described using the same reference numerals, and redundant descriptions of identical components will be omitted.

[0033] FIG. 1 is a flowchart of a solar panel monitoring method according to one embodiment of the present invention.

[0034] As illustrated in FIG. 1, a solar panel monitoring method according to one embodiment of the present invention includes a step of designating an unmanned aerial vehicle shooting point (S10), a step of operating an unmanned aerial vehicle reconnaissance route (S20), a step of acquiring unmanned aerial vehicle solar panel image information (S30), and a step of analyzing unmanned aerial vehicle solar panel image information (S40).

[0035] In the unmanned aerial vehicle shooting point designation step (S10), the flight path where the solar panels requiring investigation are located is input into the unmanned aerial vehicle to set the automatic flight path of the unmanned aerial vehicle.

[0036] In the unmanned aerial vehicle reconnaissance route operation step (S20), the investigator operates the unmanned aerial vehicle (UAV) so that it flies automatically toward the shooting point set in the unmanned aerial vehicle shooting point designation step (S10).

[0037] In the unmanned aerial vehicle solar panel image information acquisition step (S30), the unmanned aerial vehicle takes a picture at a preset altitude from the unmanned aerial vehicle shooting point to acquire image information.

[0038] In the unmanned aerial vehicle solar panel image information analysis step (S40), the image information acquired in the unmanned aerial vehicle solar panel image information acquisition step (S30) is received as thermal image information and processed using video mosaic to monitor the malfunction status of the solar panel through the distribution ratio of the solar cell surface temperature. This unmanned aerial vehicle solar panel image information analysis step (S40) will be explained in more detail in the following description.

[0040] Hereinafter, an experimental area, experimental conditions, experimental apparatus, and a data collection method for testing a solar panel monitoring method according to one embodiment of the present invention will be described.

[0041] First, referring further to FIG. 2, FIG. 2 is a photograph of a solar panel taken using a solar panel monitoring method according to one embodiment of the present invention.

[0042] As shown in Fig. 2, an experiment was conducted by photographing three solar cells with different dust distributions located at Bokhyun Hall, Kyungpook National University, Daegu, southeastern South Korea, using an unmanned aerial vehicle. In this case, the three solar cells shown in Fig. 2 were set to be covered with dust at 50%, 5%, and 0.1%, respectively. Additionally, the footage captured by the unmanned aerial vehicle was taken on August 6, 2021, when the highest solar zenith angle (13:00) occurred, in order to avoid shade and adverse weather conditions (e.g., rain). The unmanned aerial vehicle used for the recording was a DJI Matrice 200 V2 quadcopter equipped with a Zenmuse XT2 camera, as shown in Table 1. In this case, the specifications of the solar cells were Canadian Solar HiDM 400W, CS1U-400MS, 992 × 2078 mm.

[0043]

[0044] Meanwhile, the three solar cells are single photovoltaic modules composed of 162 solar cells and are assembled in a shape similar to the roof tile pattern of a building. It has been explained by Chaichan and Kazem (2020) that the illuminance of such solar cells decreases depending on the amount of accumulated dust, and specifically, it was explained that if the transmittance decreases by more than 2%, the average productivity (voltage, current, power) of the solar cells is significantly reduced. In addition, while the decrease in solar cell productivity efficiency is relatively stable at a transmittance of 2 to 20%, if the transmittance exceeds 20%, the solar cell productivity efficiency continues to decrease.

[0045] Accordingly, as described above, the testbed was classified into three types of solar cells, Testbed 1, Testbed 2, and Testbed 3, and the experiment was conducted by setting the radiation amount differently for each solar cell.

[0046] Meanwhile, to detect thermal defects in the solar cell surface temperature (SCST), the measurement distance of the ground sampling distance (GSD) of the thermal still image and video frame must be less than 16 cm.

[0047] Accordingly, the ground sampling distance (GSD) of the unmanned aerial vehicle (UAV) image in the unmanned aerial vehicle solar panel image information analysis step (S40) is the width (Sw) of the camera sensor, the flight height (H), and the focal length (F) of the camera. R ), and the image width (imW) was estimated as follows (1), and this equation (1) is applied to the unmanned aerial vehicle solar panel image information analysis step (S40).

[0048]

[0049] In this case, since the camera specifications are fixed, flight height is a key factor in determining the ground sampling distance (GSD). For this reason, by setting the flight altitude of the unmanned aerial vehicle to 30 m to leave sufficient space for the unmanned aerial vehicle to fly freely over the solar cells, a ground sampling distance (GSD) of 2.68 cm can be achieved to detect heat deficiency for each individual module of the solar cells in the experimental area.

[0050] Accordingly, the thermal infrared (TIR) ​​video in the form of a sequence format (SEQ) file captured by the unmanned aerial vehicle is captured at 30 frames per second when flying at a flight speed of 2.7 m / s at a flight altitude of 30 m and applied to the unmanned aerial vehicle solar panel image information analysis step (S40).

[0051] In this case, the thermal video captured by the unmanned aerial vehicle itself does not contain geometric information. However, the unmanned aerial vehicle (specification: DJI Matrice 200 V2) and the camera (specification: Zenmuse XT2) provide telemetry data for the entire direction (position and altitude) in a subtitle format (SubRip subtitle: SRT) along with the recorded thermal video. Additionally, the time synchronization function of OSDK V3.8.1 embedded in the flight controller of the unmanned aerial vehicle (specification: DJI Matrice 200 V2) aligns the video observation time, GPS time, and the flight controller clock to 1 Hz. Therefore, the SRT file provides the entire direction data in seconds, along with numbers representing the sequence acquired by the flight controller during flight, Coordinated Universal Time (UTC), and the entire direction parameters (GPS coordinates, barometric altitude), and is applied and used in the unmanned aerial vehicle solar panel image information analysis step (S40). In addition, in the unmanned aerial vehicle solar panel image information analysis step (S40), a thermal video frame is used for thermal image analysis, and in this case, the thermal video frame is extracted from the SEQ video file using the thermal image analysis software Thermal FLIR Studio PRO.

[0052] In this case, the total frame rate of the unmanned aerial vehicle (specification: DJI Matrice 200 V2) is 30Hz (30 frames per second), and as mentioned in Hwang et al (2021), because the frame interval is short (15 frames / 2s), the thermal signal obtained from the video mosaic is well aligned with the photo mosaic. Therefore, the frame interval for creating the video mosaic is set to 15 frames / 2 seconds to achieve 99% overlap. Additionally, automatic flight is executed via a dual grid path with a 95% overlap rate.

[0053] In addition, in the unmanned aerial vehicle solar panel image information analysis step (S40), the captured images are analyzed by video-mosaic processing, and at this time, each image captured by the unmanned aerial vehicle is automatically processed using Pix4D Mapper, a photogrammetry software, for mosaic processing.

[0054] In this case, the unmanned aerial vehicle solar panel image information analysis step (S40) proceeds by including an initial processing step, a point cloud and mesh generation step, and a digital surface model (DSM) development step in order to analyze the captured images by video mosaic processing. First, the initial processing step includes a key point extraction process for images within video frames acquired through the unmanned aerial vehicle, a key point matching process, a camera model optimization process, and a process for acquiring the geographic location GPS / GCP of the shooting point. Next, in the point cloud and mesh generation step, the points of the images within the video frames acquired at the shooting point after the initial processing step are converted into a 3D texture mesh in a high-density state. Next, in the digital surface model (DSM) development step, the images determined by the point cloud and mesh generation step are each indexed and then video mosaic processed.

[0055] Meanwhile, in the unmanned aerial vehicle solar panel image information analysis step (S40), geometric correction is performed on the captured image. More specifically, it is based on the Structure-from-Motion (SfM) algorithm, which is one of the most widely used methods for geometric correction of images captured by an unmanned aerial vehicle. This SfM algorithm enables the generation of an accurate 3D model from the image structure without prior information regarding the position or camera parameters at the time of image acquisition. In this embodiment, by using the Structure-from-Motion (SfM) algorithm, it is possible to reconstruct a 3D scene shape and camera motion from a sequence of 2D images captured by a camera of an unmanned aerial vehicle moving around.

[0056] In this case, the Structure-from-Motion (SfM) algorithm detects 16 common feature points across multiple images and uses them to reconstruct the movement of the points throughout the entire image sequence. Through this reconstruction information, the location of the points can be calculated and visualized as a 3D point cloud.

[0057] In addition, the Structure-from-Motion (SfM) algorithm is configured by applying camera exposure positions and motion trajectories to construct a sparse point cloud, and the sparse point cloud is used for camera correction and a dense point cloud is constructed by utilizing Multi-View Stereo (MVS) along with a DSM generation method based on inverse distance weighted interpolation.

[0058] In the manner described above, when the unmanned aerial vehicle solar panel image information analysis step (S40) is performed to analyze the video frame, a video mosaic as shown in FIG. 3 is formed.

[0059] Here, FIG. 3(a) is a photo-mosaic with multiple mosaic images superimposed in a thermal mosaic of an experimental site with a point cloud, and FIG. 3(b) is a video-mosaic with multiple mosaic images superimposed in a thermal mosaic of an experimental site with a point cloud.

[0060] In this case, the green area in Figures 3(a) and 3(b) indicates the degree of overlap between five or more images, while the red and yellow areas indicate low overlap and poor mosaic quality.

[0061] Here, looking at the green area of ​​Fig. 3, five or more images are superimposed, and excluding the outer border of the green area, in the case of both photo-mosaic and video-mosaic, five or more images are mostly superimposed to form.

[0062] In this case, the overlap ratio, key points, and matching key points appear as high-quality results, as shown in Fig. 3.

[0063] Based on these results, further refer to Table 2, which shows the solar cell surface temperature (SCST) in photo-mosaic and video-mosaic.

[0064]

[0065] As shown in Table 2, it can be seen that the solar cell surface temperature values ​​are nearly similar for the still image photo-mosaic and the video frame image video-mosaic.

[0066] Thus, unlike conventional methods that use high-resolution still images for photo-mosaic analysis, the solar panel monitoring method according to one embodiment of the present invention enables accurate analysis of the solar cell surface temperature even when analyzing the video-mosaic of video frames.

[0068] The following is an explanation of how to perform hotspot analysis using spatial autocorrelation.

[0069] Spatial autocorrelation is based on the first law of geography, which was mentioned by Tobler (1970), that "everything is related, but closer is more related than farther." In this case, the spatial autocorrelation technique Getis-ord Gi* is used. Getis-ord Gi* is a technique that statistically measures the degree of spatial autocorrelation to test the null hypothesis that there are no spatial patterns other than a random distribution in the region of interest.

[0070] In this case, the experiment explored the proximity distances and patterns of solar cells malfunctioning due to dust within the photo-mosaic and video-mosaic using hotspot analysis with ArcMap 9.3, a software for spatial information analysis. Accordingly, after hotspot analysis, the distribution of solar cell surface temperatures at the corresponding locations was evaluated to determine whether the spatial pattern of solar cell surface temperatures appeared identically in the photo-mosaic and video-mosaic. In this embodiment, the Getis-Ord Gi* statistic was calculated using spatial weights of the solar cells.

[0071] As a result, spatial clustering is determined through the high and low values ​​of the calculated p-value and z-score.

[0072] In this case, the formula for calculating the Getis-Ord Gi* statistics is as shown in Equation (2), Equation (3), and Equation (4).

[0073]

[0074] Here, x j are the solar cell surface temperatures for point j, and w i,jε is the spatial weight between points i and j, and n is the total solar cell surface temperature. The Getis-Ord Gi* statistic, calculated by considering the distance between points, determines statistical significance through a z-test. In other words, the degree of concentration of the upper and lower limits of the solar cells is determined through statistical significance. If the z-score is positive (+), it is determined that high values ​​are spatially clustered, and if the z-score is negative (-), it is determined that low values ​​are spatially clustered.

[0076] Next, we will describe ordinary least squares (OLS) linear regression for evaluating the performance of video mosaics.

[0077] Linear regression assumes a fixed relationship across the study area, and in this case, linear regression and Pearson correlation analysis can explain the linear relationship between the two variables based on a proportional equation.

[0078] Comparative evaluation using linear regression and Pearson correlation can yield the suitability of solar cells due to dust detected in photo-mosaics and video-mosaics.

[0079] In this case, the coefficient and error terms are used to determine whether the solar cell surface temperature obtained from the video mosaic is over- or under-measured compared to the solar cell in the photo mosaic when a heat deficiency in the solar cell is detected. Pearson correlation indicates that the solar cell surface temperatures of the video mosaic and the photo mosaic have the same orientation within the point of malfunction caused by dust.

[0080] Given that the coefficient and error terms are close to 1 and 0, respectively, the higher the Pearson correlation value, the more similar the solar cell surface temperatures obtained from photo-mosaic and video-mosaic will be in the solar heat deficiency test of a solar cell malfunctioning due to dust.

[0081] Therefore, the corresponding linear regression model is established using the solar cell surface temperature at the malfunction point detected due to dust in the video mosaic (explanatory variable) and photo mosaic (dependent variable).

[0082] This linear model is expressed by equations (5) and (6) and is corrected using ordinary least squares (OLS) estimation.

[0083]

[0084] SCST here p1 is the surface temperature of the hotspot solar cell in Testbed 1 detected in the photo-mosaic, and SCST p2 is the surface temperature of the hotspot solar cells of Testbed 2 and Testbed 3 detected in the photo-mosaic, and SCST v1 is the surface temperature of the hotspot solar cell of Testbed 1 detected in the video, is a mosaic, and SCST v2 is the surface temperature of the hotspot solar cells of Testbed 2 and Testbed 3 detected in the video mosaic, a1 and b1 are coefficients, and ε1 and ε2 are arbitrary error terms.

[0085] Accordingly, the solar panel monitoring method according to one embodiment of the present invention enables accurate analysis of the solar cell surface temperature even by analyzing the video mosaic of a video frame.

[0087] Below, histogram-based correlation will be explained to evaluate the performance of video mosaics.

[0088] Histogram-based correlation is generally used to validate feature extraction or image segmentation of specific targets by comparing threshold values ​​and distribution similarity between experimental and control groups.

[0089] For example, histogram-based correlations can reveal differences in characteristics between the control group (photo-mosaic) and the experimental group (video-mosaic) caused by thermal depletion of solar cells due to dust.

[0090] In this case, histogram-based correlation is performed using MATLAB R2021a to evaluate the distribution characteristics within solar cell hotspots caused by dust between the control group (photo-mosaic) and the experimental group (video-mosaic).

[0091] In this case, since the sample is not bivariate—that is, measurement is not performed simultaneously with observation—measurement is made based on the histogram.

[0092] For example, X i and Y j is distributed across N equal-distance bins according to its size, resulting in a relative frequency f n and g n Histogram-based correlations assuming , n = 1....,N are calculated by the following equation (7).

[0093]

[0094] Consequently, a single video frame has a lower resolution than a still photograph. However, as shown in , a video frame is compared to a still photograph (14.1–26.2 / m 3 Compared to ), key points per cubic meter (48.5~53.4 / m²) in the process of making a mosaic 3 There are more ).

[0095]

[0096] In addition, the video frame shows 110 to 110 key points per solar module (162 solar cells), which is much more than the still photo, which has 29 to 54 key points per solar module.

[0097] Here, further referring to FIG. 4, FIG. 4(a) is the differentiated distribution density of matched keypoints for a photo-mosaic of a solar cell, and FIG. 4(b) is the differentiated distribution density of matched keypoints for a photo-mosaic of a solar cell.

[0098] As shown in Figure 4, it can be seen that video mosaics have about 2.5 to 3.4 times more matching key points per solar module than photo mosaics.

[0099] In this case, since the autopilot of a conventional unmanned aerial vehicle is set to take photos over a wide area at a higher altitude, only three still photos can be used to create a mosaic from the surveyed solar panel. In contrast, the solar panel monitoring method according to an embodiment of the present invention can obtain more matched key points due to the constant and continuous superposition of 24 video frames acquired from the solar panel.

[0100] Accordingly, the solar panel monitoring method according to one embodiment of the present invention enables securing a larger number of matched key points than a photo mosaic when analyzing a video mosaic of a video frame.

[0102] Meanwhile, referring further to FIG. 5, FIG. 5(a1) is an overlaid on-field photograph of a hotspot boundary (derived from a photo-mosaic) of a solar cell malfunctioning due to dust, FIG. 5(b1) is an overlaid on-field photograph of a hotspot boundary (derived from a video-mosaic) of a solar cell malfunctioning due to dust, FIG. 5(a2) is a photograph of the hotspot boundary detected in the photo-mosaic for FIG. 5(a1), and FIG. 5(b2) is a photograph of the hotspot boundary detected in the video-mosaic for FIG. 5(b1).

[0103] In this case, Table 4 is a table of hotspot analysis results (Getis-Ord Gi) for Figure 5, and in this case, similarity is calculated by dividing the hotspots by the number of matched hotspots between the photo-mosaic and the video-mosaic.

[0104]

[0105] Referring further to FIG. 6, FIG. 6(a1) is an enlarged view of a portion of the scene photo detected in the photo-mosaic of FIG. 5(a1), FIG. 6(b1) is an enlarged view of a portion of the scene photo detected in the video-mosaic of FIG. 5(b1), FIG. 6(a2) is an enlarged view of a hotspot boundary of a portion of FIG. 6(a1), and FIG. 6(b2) is an enlarged view of a portion of FIG. 6(b1).

[0106] The hotspot analysis results for Figure 6 are statistically analyzed using Getis-Ord Gi and are shown in Table 5.

[0107]

[0108] First, Figure 5 and Table 4 show the hotspot analysis results of the surface temperatures of the solar cells in testbeds 1 to 3 between photo-mosaic and video-mosaic.

[0109] In this case, the average z-score is lower in Testbed 3 (where dust covers less than 1% of the solar cell surface) and higher in Testbed 1 (where dust covers more than 80% of the solar cell surface) compared to other testbeds in both photo-mosaic and video-mosaic.

[0110] In addition, as shown in Fig. 5, the hotspot analysis results of the solar cell surface temperature indicate that the solar cell surface temperature values ​​are high in the solar cells covered with dust in both the photo-mosaic and the video-mosaic, and the video-mosaic shows an average z-score (2.04-3.03) similar to the photo-mosaics (2.22-3.14) of testbeds 1 to 3.

[0111] This means that the video mosaic has a spatial pattern similar to the hotspots derived from the photo mosaic. The spatial pattern similarity of the hotspots is over 90% between the photo mosaic and the video mosaic in testbeds 1 and 3, while testbed 2 (soil residue accounting for less than 10-20% of the solar modules) shows the lowest similarity at 81%.

[0112] Figure 6 shows enlarged sections of the photo-mosaic and video-mosaic with hotspot boundaries, as well as field photographs of Test Bed 2. In this case, cases where the solar cell surface temperature is measured to be high are mostly clearly visible in the video-mosaic of the test bed covered with dust. That is, as shown in Table 4, the higher the solar cell surface temperature value, the more scattered it is in the video-mosaic than in the photo-mosaic.

[0113] Table 5 and Figure 7 show the results of regression analysis to investigate the solar cell surface temperature at which a common hotspot was detected in the video-mosaic and photo-mosaic of the testbed.

[0114] In this case, SCST v1 and SCST v2 Since the Pearson correlation coefficient is higher than 0.98 (0.981–0.991), SCST p1 and SCST p2 It shows a strong correlation with. In addition, the model fit value is very high, ranging from 0.981 to 0.962.

[0115] In addition, the unstandardized coefficients representing the direction and intensity of solar cell surface temperature (SCST) detected in photo-mosaic and video-mosaic range from 0.973 to 1.003, showing a strong positive (+) linear correlation.

[0116] This implies that the increase in solar cell surface temperature caused by defective solar cells may point in the same direction.

[0117] In contrast, the models of Testbeds 2 and 3 have lower unstandardized factors than Testbed 1. In this case, referring further to Fig. 7, Fig. 7(a) is a scatter plot of solar cell surface temperature (SCST) at the matched hotspot between the photo-mosaic and video-mosaic for Testbed 1, and Fig. 7(b) is a scatter plot of solar cell surface temperature (SCST) at the matched hotspot between the photo-mosaic and video-mosaic for Testbeds 2 and 3. Therefore, as shown in Fig. 7, SCST v2 SCST p2 It indicates that SCSTS is higher than

[0118] Similarly, similar results can be found when testing the null hypothesis that the coefficients (a1, a2) of individual models are 0.

[0119] To test the null hypothesis accordingly, t-statistics and p-values ​​were calculated. The p-values ​​indicated the highest error probability where H0 could not be ignored, leading to the conclusions that "there is no difference between the two distributions" and "there is a statistically significant difference between the two distributions."

[0120] Here, since all p-values ​​are less than 0.05, H0 cannot be ignored based on all data.

[0121] The t-statistic cast ε₀ is the sample mean, S is defined as the standard deviation, and N represents the sample size; generally, a higher t-statistic value indicates that H₀ may be neglected. In other words, the joint coefficient is more likely to be significant. The precise boundary between significance and insignificance depends on the sample size and the Student's t-distribution. In Testbed 2 and Testbed 3, the t-statistic values ​​were measured to be lower than in Testbed 1, ranging from 199.847 to 46.949. The RMSE in Testbed 2 and Testbed 3 was slightly higher than in Testbed 1, ranging from 0.136°C to 0.183°C. However, the benchmark for the t-statistic is generally lower than in Testbed 3. Therefore, the coefficient is statistically significant given the t-statistic values ​​in Table 5.

[0122] The similarity in the distribution of solar cell surface temperatures at the matched hotspots between the photo-mosaic and video-mosaic of solar cells malfunctioning due to dust can be presented as a distribution characteristic of the observed records of Testbed 1 and Testbeds 2 and 3.

[0123] In this case, a time record difference of approximately 30 minutes, sufficient to distort the measured temperature, must be explained; to do this, the two observations must be averaged. Accordingly, the i-th temperature value observed in the photo-mosaic is auto i Represented as , and the i-th temperature value observed in the video mosaic is auto j It is represented as.

[0124] Then, ...comes to consider.

[0125] In this case, there are various options for comparing distribution properties, and most methods assume that the underlying observations follow a Gaussian distribution.

[0126] In this case, referring further to Fig. 8 and Table 6, Fig. 8(a) is a comparison graph of the solar cell surface temperature adjustment values ​​of the photo-mosaic (yellow bar) and the solar cell surface temperature adjustment values ​​of the video mosaic (blue bar) for test bed 1, and Fig. 8(b) is a comparison graph of the solar cell surface temperature adjustment values ​​of the photo-mosaic (yellow bar) and the solar cell surface temperature adjustment values ​​of the video mosaic (blue bar) for test beds 2 and 3.

[0127] Table 6 is a descriptive statistical table of histograms for the average adjusted solar cell surface temperature (SCST) at the matched hotspot between the photo-mosaic and video-mosaic.

[0128]

[0129] Referring to the histogram in Figure 8 and the statistical values ​​in Table 6, it can be seen that all distributions are heavily skewed to the right.

[0130] For n = 20 in Testbed 1, ρ hist A histogram correlation of 98.95% was observed, and for Testbeds 2 and 3, the value was ρ hist It was found to be slightly lower at 90.52%.

[0131] Since this value varies depending on the number of bins but the magnitude remains the same, interpretation: a very high correlation value close to +1 can be observed, indicating that the two histograms are highly correlated, i.e., significantly identical.

[0132] In this way, the solar panel monitoring method according to one embodiment of the present invention enables an unmanned aerial vehicle to photograph a solar cell covered with dust using automatic control, and to automatically detect a solar cell malfunctioning due to dust by automatically analyzing the captured image.

[0133] In particular, the solar panel monitoring method according to one embodiment of the present invention enables inspection at the solar panel level, thereby allowing for more sophisticated analysis than conventional methods, as it overcomes the problem of not being able to perform inspection at the solar panel level due to a lack of key points and unnecessary data duplication in conventional solar cell thermal defect inspection methods. This is achieved by performing inspection at the solar panel level, rather than analyzing only the image of the entire area where the solar panels are installed.

[0134] In addition, the solar panel monitoring method according to one embodiment of the present invention enables the detection of spatial patterns and distribution characteristics at the level of a photo-mosaic using still images, even when using video frames with a resolution lower than that of still images when detecting spatial patterns and distribution characteristics.

[0135] In addition, since the solar panel monitoring method according to one embodiment of the present invention uses video mosaic, it can find 2.5 to 3.4 times more matching key points per solar module than photo mosaic, thereby providing a superior analysis method compared to conventional methods.

[0137] As described above, the present invention has been explained by specific details such as specific components, limited embodiments, and drawings; however, this is provided merely to aid in a more comprehensive understanding of the invention, and the invention is not limited to the above embodiments. A person skilled in the art can make various modifications and variations from this description.

[0138] Accordingly, the scope of the present invention should not be limited to the described embodiments, and all things equivalent to or having equivalent variations to the claims set forth below, as well as the claims set forth below, shall be considered to fall within the scope of the concept of the present invention.

Claims

Claim 1 A solar panel monitoring method comprising: an image information acquisition step in which an unmanned aerial vehicle performs shooting at a preset altitude to acquire image information regarding a solar panel; and an image information analysis step in which the image information is received as thermal image information and analyzed for a surface temperature distribution rate regarding the solar panel by video-mosaic processing; wherein the image information acquisition step further comprises acquiring the image information as continuously captured video frames; and wherein the image information analysis step further comprises: an initial processing step including a key point extraction process, a key point matching process, a camera model optimization process, and a GPS / GCP acquisition process for each of the images within the video frames acquired through the unmanned aerial vehicle; a point cloud and mesh generation step in which the points of the images within the video frames acquired at the shooting point after the initial processing step are converted into a 3D texture mesh in a high-density state; and a digital surface model (DSM) development step in which the images determined by the point cloud and mesh generation step are each indexed and then video-mosaic processed. Claim 2 A solar panel monitoring method according to claim 1, further comprising, in the image information analysis step, analyzing the surface temperature distribution rate for the cell unit of the solar panel. Claim 3 A solar panel monitoring method according to claim 1, further comprising, in the image information analysis step, detecting a spatial pattern during the video-mosaic processing. Claim 4 delete Claim 5 In claim 1, in the image information analysis step, the ground sampling distance (GSD) of the image captured by the unmanned aerial vehicle is given by Equation (1). It is determined as, where Sw is the width of the camera sensor, H is the flight height, and F R A solar panel monitoring method comprising: a focal length of a camera, imW being the image width. Claim 6 A solar panel monitoring method according to claim 1, further comprising: geometrically correcting the captured image based on a Structure-from-Motion (SfM) algorithm in the image information analysis step. Claim 7 A solar panel monitoring method according to claim 6, further comprising: reconstructing geometric data into a 3D scene shape and camera motion using a 2D image sequence captured by a camera of an unmanned aerial vehicle when using the SfM (Structure-from-Motion) algorithm. Claim 8 A solar panel monitoring method according to claim 1, further comprising: in the image information analysis step, generating spatial information about time using an SRT file that provides total direction data in seconds along with numbers representing sequences, Coordinated Universal Time (UTC), and total direction parameters acquired from the flight controller of the unmanned aerial vehicle. Claim 9 delete

Images