A picture recognition result screening method, device and equipment and storage medium

Abstract

The application relates to the technical field of image recognition, and particularly discloses a picture recognition result screening method and device, equipment and a storage medium. The method comprises the following steps: acquiring a cropped subpicture, the cropped subpicture containing at least one photovoltaic panel; inputting the cropped subpicture into a photovoltaic panel inference model to obtain coordinate information corresponding to at least one prediction box; calculating the contour area of the prediction box corresponding to each coordinate information, and sorting the contour areas to obtain a prediction box area sequence; traversing the coordinate information corresponding to each prediction box in the prediction box area sequence to obtain an associated prediction grouping of all prediction boxes containing corresponding photovoltaic panels; and selecting the prediction box with the largest contour area in each associated prediction grouping as a screening result of a target photovoltaic panel. Through area sorting and associated grouping, all prediction boxes of the same photovoltaic panel are collected, the prediction box with the largest area is selected as the result, the actual contour of the photovoltaic panel is accurately reflected, and the photovoltaic panel recognition accuracy is improved.

Description

Technical Field

[0001] This application relates to the field of image recognition technology, and in particular to a method, apparatus, device and storage medium for filtering image recognition results. Background Technology

[0002] For deep learning models like YOLOv, to ensure the accuracy of model inference, it is necessary to limit the pixel size of the images input to the model (common image sizes are mainly 640px and 1280px). However, this is crucial for surveying photovoltaic power plants. Figure 1 These maps are generally quite large, with pixel sizes typically exceeding 10,000 pixels. If such maps are directly input into the inference model, the recognition results will be unsatisfactory.

[0003] For reasoning with such large images, a slicing technique is needed. A large power plant survey map is cropped into smaller images and input into the model for recognition. Since the cropping location is random, it is inevitable that some photovoltaic panels will be cut off, and the model may also recognize these cut-off parts of the photovoltaic panels. Therefore, a slicing algorithm needs to be designed to accurately filter out the predicted bounding boxes of the cut-off photovoltaic panels, retaining only the complete predicted bounding boxes. Summary of the Invention

[0004] The main purpose of this application is to provide a method for filtering image recognition results, which aims to solve the technical problem of low accuracy in filtering photovoltaic panel prediction frames in the prior art.

[0005] To achieve the above objectives, this application provides a method for filtering image recognition results, the method comprising the following steps: Obtain a cropped sub-image, wherein the cropped sub-image contains at least one photovoltaic panel; The cropped sub-image is input into the photovoltaic panel inference model to obtain the coordinate information of at least one prediction box. The prediction box is a rectangular box that selects each outline of the photovoltaic panel by the photovoltaic panel inference model. Calculate the outline area of ​​the prediction box corresponding to each of the coordinate information, and sort the outline areas to obtain the prediction box area sequence; By traversing the coordinate information corresponding to each prediction box in the prediction box area sequence, the associated prediction group of each photovoltaic panel is obtained, and each associated prediction group contains all the prediction boxes of the corresponding photovoltaic panel. The prediction box with the largest outline area in each of the associated prediction groups is selected as the screening result of the target photovoltaic panel.

[0006] In one embodiment, the step of obtaining the cropped sub-image includes: Obtain the dimensional parameters of the photovoltaic panel; The overlap rate between adjacent sub-graphs is set according to the size parameters of the photovoltaic panel; The photovoltaic panel's survey image is cropped based on the overlap ratio to obtain a cropped sub-image.

[0007] In one embodiment, the step of inputting the cropped sub-image into the photovoltaic panel inference model to obtain coordinate information corresponding to at least one prediction box includes: Determine whether the cropped sub-image contains a prediction box; If not, the first cropped sub-image that does not contain the prediction box is input into the photovoltaic panel inference model to perform photovoltaic panel recognition inference, and a second cropped sub-image containing at least one prediction box is generated. The second cropped sub-image contains the prediction boxes corresponding to all the photovoltaic panels. Calculate the coordinate information corresponding to each of the prediction boxes.

[0008] In one embodiment, the step of calculating the contour area of ​​the predicted bounding box corresponding to each of the coordinate information, and sorting the contour areas to obtain a sequence of predicted bounding box areas, includes: The vertex coordinates of the corresponding prediction box are obtained based on the coordinate information provided. Calculate the contour area based on the vertex coordinates; The predicted box area sequence is obtained by sorting the outline areas in descending order.

[0009] In one embodiment, the step of traversing the coordinate information corresponding to each prediction box in the prediction box area sequence to obtain the associated prediction grouping of each photovoltaic panel includes: Based on the order of the contour area from largest to smallest, one of the predicted boxes is taken out sequentially from the predicted box area sequence as the box to be classified. Calculate the difference between the coordinate information corresponding to the box to be classified and the coordinate information corresponding to the remaining predicted boxes in the predicted box area sequence; The prediction frames whose differences are below a preset error range are assigned to the associated prediction groups corresponding to the same photovoltaic panel.

[0010] In one embodiment, the coordinate information includes: contour coordinates and center point coordinates, and the step of calculating the difference between the coordinate information corresponding to the box to be classified and the coordinate information corresponding to the remaining predicted boxes in the predicted box area sequence includes: The area overlap rate is calculated based on the first contour coordinates corresponding to the box to be classified and the second contour coordinates corresponding to the remaining predicted boxes in the predicted box area sequence. The center point distance is calculated based on the first center point coordinates corresponding to the box to be classified and the second center point coordinates corresponding to the remaining predicted boxes in the predicted box area sequence. The difference is obtained based on the area overlap rate or the center point distance.

[0011] In one embodiment, the step of assigning prediction frames with differences below a preset error range to the associated prediction groups corresponding to the same photovoltaic panel includes: Remove predicted boxes whose difference is below a preset error range from the predicted box area sequence; The difference comparison is performed sequentially until each prediction box in the prediction box area sequence is assigned to the corresponding associated prediction group.

[0012] Furthermore, to achieve the above objectives, this application also provides an image recognition result filtering device, the device comprising: An image acquisition module is used to acquire a cropped sub-image, wherein the cropped sub-image contains at least one photovoltaic panel; The coordinate recognition module is used to input the cropped sub-image into the photovoltaic panel inference model to obtain the coordinate information of at least one prediction box. The prediction box is a rectangular box that selects each outline of the photovoltaic panel and is recognized and output by the photovoltaic panel inference model. An area sorting module is used to calculate the contour area of ​​the prediction box corresponding to each of the coordinate information, and sort the contour areas to obtain a prediction box area sequence. The prediction grouping module is used to traverse the coordinate information corresponding to each prediction box in the prediction box area sequence to obtain the associated prediction group of each photovoltaic panel. Each associated prediction group contains all the prediction boxes of the corresponding photovoltaic panel. The result recognition module is used to select the prediction box with the largest outline area in each of the associated prediction groups as the screening result of the target photovoltaic panel.

[0013] In addition, to achieve the above objectives, this application also provides an image recognition result filtering device, which includes: a memory, a processor, and an image recognition result filtering processing program stored in the memory and executable on the processor. When the image recognition result filtering processing program is executed by the processor, it implements the steps of the above-described image recognition result filtering method.

[0014] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the image recognition result filtering method described above.

[0015] The above-mentioned one or more technical solutions provided in this application may have the following advantages or at least achieve the following technical effects: This application discloses a method, apparatus, device, and storage medium for filtering image recognition results. The method includes: acquiring a cropped sub-image containing at least one photovoltaic panel; inputting the cropped sub-image into a photovoltaic panel inference model to obtain coordinate information corresponding to at least one predicted bounding box; calculating the contour area of ​​the predicted bounding box corresponding to each coordinate information, and sorting the contour areas to obtain a predicted bounding box area sequence; traversing the coordinate information corresponding to each predicted bounding box in the predicted bounding box area sequence to obtain an associated predicted group containing all predicted bounding boxes of the corresponding photovoltaic panel; and selecting the predicted bounding box with the largest contour area in each associated predicted predicted group as the filtering result for the target photovoltaic panel. By sorting and associating the predicted bounding boxes by area, all predicted bounding boxes of the same photovoltaic panel are aggregated, and the predicted bounding box with the largest area is selected as the result, accurately reflecting the actual contour of the photovoltaic panel and improving the accuracy of photovoltaic panel recognition. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the first embodiment of the image recognition result filtering method of this application; Figure 2 This is a cropping diagram of the first embodiment of the image recognition result filtering method of this application; Figure 3 This is a flowchart illustrating the second embodiment of the image recognition result filtering method of this application; Figure 4 This is a schematic diagram of the first position in the second embodiment of the image recognition result filtering method of this application; Figure 5 This is a schematic diagram of the second position in the second embodiment of the image recognition result filtering method of this application; Figure 6 This is a schematic diagram of the third position in the second embodiment of the image recognition result filtering method of this application; Figure 7 This is a flowchart illustrating the third embodiment of the image recognition result filtering method of this application; Figure 8 This is a schematic diagram of the module structure of the image recognition result filtering device of this application; Figure 9 This is a schematic diagram of the hardware operating environment for the image recognition result filtering device involved in the image recognition result filtering method of this application.

[0018] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0020] It should be noted that if the embodiments of this application involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a specific posture. If the specific posture changes, the directional indicators will also change accordingly.

[0021] Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the use of "and / or" or "and / or" throughout the text includes three parallel solutions. For example, "A and / or B" includes solution A, solution B, or a solution that simultaneously satisfies A and B. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0022] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an image recognition result filtering device capable of performing the above functions. The following description uses an image recognition result filtering device as an example to illustrate this embodiment and the subsequent embodiments.

[0023] Understandably, the commonly used Sahi slice inference algorithm employs Non-Maximum Suppression (NMS), which selects the prediction box with the highest confidence. However, the confidence level is easily affected by environmental factors, and the confidence level of a prediction box that truncates the photovoltaic panel may not be lower than that of a prediction box that is complete. In such cases, using the NMS algorithm for filtering will cause problems.

[0024] Based on this, embodiments of this application provide a method for filtering image recognition results, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the image recognition result filtering method of this application.

[0025] Step S10: Obtain the cropped sub-image.

[0026] It should be noted that the cropped sub-image includes at least one photovoltaic panel. A photovoltaic panel can be a power generation unit of a photovoltaic power station, capable of converting light energy into electrical energy for storage and transmission. A photovoltaic power station can contain many photovoltaic panels connected in series or parallel, which are the target objects to be identified and screened in this application. The survey image of the photovoltaic panel can be a very large-pixel aerial survey image of the photovoltaic power station, with a pixel size generally exceeding 10,000, making it impossible to directly input into a deep learning model for analysis and reasoning.

[0027] Understandably, by calling the API of the gdal library, a large-scale survey map can be cropped to form a cropped sub-image according to a set size and overlap rate. The cropped sub-image can be a small-sized image obtained after cropping the large-scale survey map, which can be adapted to the input pixel requirements of deep learning models, typically 640 / 1280 pixels (px), ensuring that a single photovoltaic panel is complete in at least one sub-image.

[0028] In one possible implementation, to ensure that at least one complete photovoltaic panel exists in the cropped sub-image, the following steps can be taken: obtaining the size parameters of the photovoltaic panel; setting the overlap rate between adjacent sub-images based on the size parameters of the photovoltaic panel; and cropping the measured image of the photovoltaic panel based on the overlap rate to obtain the cropped sub-image.

[0029] It should be noted that the dimensions of the photovoltaic panel can be physical dimensions such as the horizontal length and vertical width of a single photovoltaic panel. Adjacent sub-images can be two cropped sub-images that are adjacent in position and have overlapping areas after cropping an image of a large photovoltaic system. (Refer to...) Figure 2 , Figure 2 This is a schematic diagram of cropping in the first embodiment of the image recognition result filtering method of this application. Cropping is performed according to a certain cropping size and overlap rate. The overlap rate can be the ratio of the length of the overlapping area between two adjacent cropped sub-images to the side length of the sub-image. By setting the overlap length and width to be greater than the length and width of a single photovoltaic panel, it can be ensured that a single photovoltaic panel is complete in at least one sub-image. (Refer to...) Figure 2 The overlap length PQ between the three cutting frames (frame A, frame B, and frame C) should be greater than the horizontal length of the photovoltaic panel, and the overlap length MN between the three frames (frame A, frame B, and frame C) should be greater than the vertical width of the photovoltaic panel.

[0030] Step S20: Input the cropped sub-image into the photovoltaic panel inference model to obtain the coordinate information corresponding to at least one prediction box.

[0031] It should be noted that the photovoltaic panel inference model can be a target detection model that identifies photovoltaic panels in a sub-image and outputs corresponding outline boxes, similar to deep learning detection models such as the YOLOv series. The predicted bounding box can be a rectangular box output by the model after recognition, used to select the outline of the photovoltaic panel, including both complete photovoltaic panel boxes and truncated photovoltaic panel boxes. The coordinate information can be the coordinate information of the bounding box, the coordinate information of the four vertices of the rectangle, and the coordinate information of the center of the predicted box. In this embodiment, no specific limitation is made, but it can be used to locate and calculate the relevant parameters of the predicted box.

[0032] In one possible implementation, it can be determined whether the cropped sub-image contains a predicted bounding box. If not, the first cropped sub-image, which does not contain the predicted bounding box, is input into the photovoltaic panel inference model for photovoltaic panel recognition inference, generating a second cropped sub-image containing at least one predicted bounding box. The second cropped sub-image contains the predicted bounding boxes corresponding to all the photovoltaic panels. The coordinate information corresponding to each predicted bounding box is calculated.

[0033] It should be noted that the predicted bounding box sub-image can be a cropped sub-image that does not contain the predicted bounding box, which is then inferred by the model and an image with the predicted bounding box of the photovoltaic panel is generated, fully presenting the recognition results of all photovoltaic panels in the sub-image.

[0034] Understandably, the first step is to determine if a subgraph contains a bounding box. Inference is only performed on subgraphs without bounding boxes to avoid invalid computations and reduce the model's computational load. Some cropped subgraphs initially lack bounding boxes due to photovoltaic panel truncation or shading; skipping this step directly would miss photovoltaic panels in that area, leading to incomplete recognition results. Inference is performed on subgraphs without bounding boxes to ensure that all photovoltaic panels are recognized, leaving no targets unrecognized and guaranteeing high recognition coverage. This approach addresses situations where subgraphs initially lack bounding boxes due to photovoltaic panel truncation or partial shading, enhancing the model's adaptability to complex subgraphs.

[0035] Step S30: Calculate the contour area of ​​the prediction box corresponding to each of the coordinate information, and sort the contour areas to obtain the prediction box area sequence.

[0036] It should be noted that the outline area can be the actual area of ​​the rectangular box calculated from coordinate information (e.g., the vertex coordinates of the predicted box). All predicted boxes are arranged according to their outline areas to form a sequence of predicted box areas. The predicted box area sequence can be sorted from largest to smallest or from smallest to largest; no specific limitation is made in this embodiment. It is ensured that each predicted box appears exactly once in the predicted box area sequence according to certain rules.

[0037] Step S40: Traverse the coordinate information corresponding to each prediction box in the prediction box area sequence to obtain the associated prediction group of each photovoltaic panel.

[0038] It should be noted that the associated prediction group can be a collection of multiple prediction frames (complete frames, truncated frames) belonging to the same photovoltaic panel, and each associated prediction group contains all the prediction frames of the corresponding photovoltaic panel. Specifically, the attribution can be determined by correlation data (such as area overlap rate, center point distance) calculated from coordinate information. Therefore, all prediction frames can be classified into at least one associated prediction group.

[0039] Step S50: Select the prediction box with the largest outline area in each of the associated prediction groups as the screening result of the target photovoltaic panel.

[0040] It should be noted that the target photovoltaic panel can be a photovoltaic panel corresponding to an associated prediction group, i.e., the object whose optimal identification box is to be determined. The selection result can be the prediction box with the largest area in the associated prediction group, which is the optimal identification result that can completely reflect the actual outline of the photovoltaic panel.

[0041] In this embodiment, the image recognition result filtering method includes: obtaining a cropped sub-image containing at least one photovoltaic panel; inputting the cropped sub-image into a photovoltaic panel inference model to obtain coordinate information corresponding to at least one predicted box; calculating the contour area of ​​the predicted box corresponding to each coordinate information, and sorting the contour areas to obtain a predicted box area sequence; traversing the coordinate information corresponding to each predicted box in the predicted box area sequence to obtain an associated predicted group containing all predicted boxes of the corresponding photovoltaic panel; and selecting the predicted box with the largest contour area in each associated predicted group as the filtering result for the target photovoltaic panel. By sorting and associating the predicted boxes by area, all predicted boxes of the same photovoltaic panel are aggregated, and the predicted box with the largest area is selected as the result, accurately reflecting the actual contour of the photovoltaic panel and improving the accuracy of photovoltaic panel recognition.

[0042] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Please refer to Figure 3 , Figure 3 This is a flowchart illustrating the second embodiment of the image recognition result filtering method of this application. In this embodiment, step 30 includes: Step S301: Obtain the vertex coordinates of the corresponding prediction box based on the coordinate information of each.

[0043] It should be noted that the vertex coordinates can be the latitude and longitude values ​​of the four corner points of the rectangle extracted from the predicted bounding box coordinate information, and are direct parameters for calculating the contour area.

[0044] Step S302: Calculate the contour area based on the vertex coordinates.

[0045] It should be noted that the outline area is the actual area of ​​the rectangular box calculated from the coordinates of the four vertices of the predicted box (calculated based on the difference in horizontal and vertical length coordinates), which is different from the confidence score output by the traditional model. It is a quantitative indicator used to select the optimal target box.

[0046] Step S303: Arrange the predicted box area sequence in descending order based on the outline area.

[0047] It should be noted that in this embodiment, all predicted boxes are sorted in descending order after area calculation to form an ordered list, which is the predicted box area sequence.

[0048] Understandably, the prediction frame selection process primarily aims to choose the optimal prediction frame from multiple prediction frames for a single photovoltaic panel. Considering that the prediction frame should encompass the outline of the photovoltaic panel as much as possible, the prediction frame with the largest area is generally selected, while other redundant prediction frames are filtered out. The positional relationship between different prediction frames on the same photovoltaic panel due to truncation is as follows: smaller frames are contained within larger frames, as shown in the reference... Figure 4 , Figure 4 This is a schematic diagram of the first position in the second embodiment of the image recognition result filtering method of this application. The small frame intersects with the large frame, as shown in the diagram. Figure 5 , Figure 5 This is a schematic diagram of the second position in the second embodiment of the image recognition result filtering method of this application. The small frame and the large frame do not intersect, see reference. Figure 6 , Figure 6 This is a schematic diagram of the third position in the second embodiment of the image recognition result filtering method of this application.

[0049] In this embodiment, the vertex coordinates of the corresponding predicted bounding boxes are obtained based on the coordinate information; the contour area is calculated based on the vertex coordinates; and the predicted bounding box area sequence is obtained by sorting the contour areas in descending order. Calculating the contour area based on vertex coordinates avoids estimation errors. By prioritizing the display of large, complete bounding boxes in descending order, subsequent grouping prioritizes complete bounding boxes as the benchmark, reducing invalid comparisons and improving screening efficiency. Using area instead of easily influenced confidence levels is unaffected by environmental factors, accurately distinguishing between complete photovoltaic panel frames and truncated frames, reducing grouping error rates.

[0050] Based on the above embodiments of this application, in the third embodiment of this application, the same or similar content as the above embodiments can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 7 , Figure 7 This is a flowchart illustrating the third embodiment of the image recognition result filtering method of this application. In this embodiment, step 40 includes: Step S701: Select one predicted box from the predicted box area sequence in descending order of the contour area as the box to be classified.

[0051] It should be noted that the box to be classified can be the current predicted box taken out in order from the predicted box area sequence and used as the grouping benchmark. Since the predicted box area sequence is arranged in descending order, the complete photovoltaic panel box with the largest area is selected first.

[0052] Step S702: Calculate the difference between the coordinate information corresponding to the box to be classified and the coordinate information corresponding to the remaining predicted boxes in the predicted box area sequence.

[0053] It should be noted that the coordinate information includes: contour coordinates and center point coordinates. The area overlap rate is calculated based on the first contour coordinates corresponding to the box to be classified and the second contour coordinates corresponding to the remaining predicted boxes in the predicted box area sequence. The center point distance is calculated based on the first center point coordinates corresponding to the box to be classified and the second center point coordinates corresponding to the remaining predicted boxes in the predicted box area sequence. The difference is then obtained based on the area overlap rate or the center point distance.

[0054] It is understandable that the first contour coordinates and the first center point coordinates can be the contour coordinates and center point coordinates corresponding to the bounding box to be classified. The second contour coordinates and the second center point coordinates can be the contour coordinates and center point coordinates corresponding to the remaining bounding boxes to be compared and predicted in the area sequence.

[0055] Step S703: Divide the prediction frames whose difference is lower than the preset error range into the associated prediction group corresponding to the same photovoltaic panel.

[0056] It should be noted that the preset error range can be a pre-set judgment standard. If it is lower than the range, it is judged to be the prediction frame of the same photovoltaic panel.

[0057] In one specific implementation, the latitude and longitude coordinates of all photovoltaic panel prediction frames obtained from subgraph inference are read, and the area of ​​each prediction frame is calculated. The prediction frames are then sorted in descending order according to their area, generating a prediction frame area sequence S. One photovoltaic panel prediction frame 'a' is then extracted from the prediction frame area sequence S, and a new associated prediction group Ai is created. Prediction frame 'a' is removed from the prediction frame area sequence S and added to the associated prediction group Ai. Prediction frame 'a' is compared with the other prediction frames 'b' in the remaining sequence S, mainly calculating the area overlap rate and center point distance between the two photovoltaic panel prediction frames. The calculation methods for the area overlap rate and center point distance are as follows: Area overlap rate = Area of ​​the overlapping portion of two prediction boxes / Area of ​​the smaller prediction box; Center-to-center distance = straight-line distance between the center points of the two boxes; When prediction box a and other prediction boxes b satisfy the condition that the area overlap rate is greater than a threshold or the center point distance is less than a threshold, then a and b are considered to belong to the same photovoltaic panel. Prediction box b is removed from the prediction box area sequence S and added to the associated prediction group Ai. The above operation is repeated until all photovoltaic panel prediction boxes in the prediction box area sequence S are traversed. All prediction boxes of each photovoltaic panel are saved to their respective associated prediction group Ai. The first prediction box in the associated prediction group Ai is regarded as the prediction box with the largest area, that is, it is considered to best reflect the outline of the photovoltaic panel.

[0058] In this embodiment, a predicted frame is sequentially selected from the predicted frame area sequence as the frame to be classified; the difference between the coordinate information corresponding to the frame to be classified and the coordinate information corresponding to the remaining predicted frames in the predicted frame area sequence is calculated; predicted frames with a difference lower than a preset error range are assigned to the associated predicted group corresponding to a photovoltaic panel. Grouping starts from the frame with the largest area, using the complete photovoltaic panel frame as the benchmark, avoiding truncated frames dominating grouping, and significantly reducing the probability of different groups for the same panel and different panels in the same group. Based on the dual features of area overlap rate and center point distance, the confidence level, which is easily distorted, is not affected by environmental factors such as light shading, and accurately identifies multiple frames on the same panel.

[0059] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent scope of this application.

[0060] This application also provides an image recognition result filtering device, please refer to... Figure 8 The image recognition result filtering device includes: Image acquisition module 10 is used to acquire a cropped sub-image, wherein the cropped sub-image contains at least one photovoltaic panel; The coordinate recognition module 20 is used to input the cropped sub-image into the photovoltaic panel inference model to obtain the coordinate information of at least one prediction box. The prediction box is a rectangular box that selects each outline of the photovoltaic panel and is recognized and output by the photovoltaic panel inference model. The area sorting module 30 is used to calculate the outline area of ​​the prediction box corresponding to each of the coordinate information, and sort the outline areas to obtain the prediction box area sequence. The prediction grouping module 40 is used to traverse the coordinate information corresponding to each prediction box in the prediction box area sequence to obtain the associated prediction group of each photovoltaic panel. Each associated prediction group contains all the prediction boxes of the corresponding photovoltaic panel. The result recognition module 50 is used to select the prediction box with the largest contour area in each of the associated prediction groups as the screening result of the target photovoltaic panel.

[0061] The image recognition result filtering device provided in this application, employing the image recognition result filtering method in the above embodiments, can solve the technical problem of low filtering accuracy of photovoltaic panel prediction frames. Compared with the prior art, the beneficial effects of the image recognition result filtering device provided in this application are the same as those of the image recognition result filtering method provided in the above embodiments, and other technical features in the image recognition result filtering device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0062] This application provides an image recognition result filtering device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the image recognition result filtering method in the above embodiment 1.

[0063] The following is for reference. Figure 9 The diagram illustrates a structural schematic suitable for implementing an image recognition result filtering device according to embodiments of this application. The image recognition result filtering device in embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 9 The image recognition result filtering device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0064] like Figure 9As shown, the image recognition result filtering device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the image recognition result filtering device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. The communication device 1009 allows the image recognition result screening device to communicate wirelessly or wiredly with other devices to exchange data. Although the figures show image recognition result screening devices with various systems, it should be understood that implementing or having all of the systems shown is not required. More or fewer systems may be implemented alternatively.

[0065] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0066] The image recognition result filtering device provided in this application adopts the image recognition result filtering method in the above embodiments. Compared with the prior art, the beneficial effects of the image recognition result filtering device provided in this application are the same as the beneficial effects of the image recognition result filtering method provided in the above embodiments. Moreover, the other technical features in the image recognition result filtering device are the same as the features disclosed in the method of the previous embodiment, and will not be described in detail here.

[0067] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0068] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0069] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the image recognition result filtering method in the above embodiments.

[0070] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0071] The aforementioned computer-readable storage medium may be included in the image recognition result screening device; or it may exist independently and not be assembled into the image recognition result screening device.

[0072] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0073] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0074] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0075] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described image recognition result filtering method, which can solve the technical problem of low accuracy in photovoltaic panel prediction frame filtering. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the image recognition result filtering method provided in the above embodiments, and will not be repeated here.

[0076] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for filtering image recognition results, characterized in that, The image recognition result filtering method is applied to the filtering of photovoltaic panels, and the image recognition result filtering method includes the following steps: Obtain a cropped sub-image, wherein the cropped sub-image contains at least one photovoltaic panel; The cropped sub-image is input into the photovoltaic panel inference model to obtain the coordinate information of at least one prediction box. The prediction box is a rectangular box that selects each outline of the photovoltaic panel by the photovoltaic panel inference model. Calculate the outline area of ​​the prediction box corresponding to each of the coordinate information, and sort the outline areas to obtain the prediction box area sequence; By traversing the coordinate information corresponding to each prediction box in the prediction box area sequence, the associated prediction group of each photovoltaic panel is obtained, and each associated prediction group contains all the prediction boxes of the corresponding photovoltaic panel. The prediction box with the largest outline area in each of the associated prediction groups is selected as the screening result of the target photovoltaic panel.

2. The image recognition result filtering method as described in claim 1, characterized in that, The step of obtaining the cropped sub-image includes: Obtain the dimensional parameters of the photovoltaic panel; The overlap rate between adjacent sub-graphs is set according to the size parameters of the photovoltaic panel; The photovoltaic panel's survey image is cropped based on the overlap ratio to obtain a cropped sub-image.

3. The image recognition result filtering method as described in claim 1, characterized in that, The step of inputting the cropped sub-image into the photovoltaic panel inference model to obtain the coordinate information corresponding to at least one prediction box includes: Determine whether the cropped sub-image contains a prediction box; If not, the first cropped sub-image that does not contain the prediction box is input into the photovoltaic panel inference model to perform photovoltaic panel recognition inference, and a second cropped sub-image containing at least one prediction box is generated. The second cropped sub-image contains the prediction boxes corresponding to all the photovoltaic panels. Calculate the coordinate information corresponding to each of the prediction boxes.

4. The image recognition result filtering method as described in claim 1, characterized in that, The step of calculating the contour area of ​​the prediction box corresponding to each of the coordinate information, and sorting the contour areas to obtain a sequence of prediction box areas, includes: The vertex coordinates of the corresponding prediction box are obtained based on the coordinate information provided. Calculate the contour area based on the vertex coordinates; The predicted box area sequence is obtained by sorting the outline areas in descending order.

5. The image recognition result filtering method as described in claim 4, characterized in that, The step of traversing the coordinate information corresponding to each prediction box in the prediction box area sequence to obtain the associated prediction grouping of each photovoltaic panel includes: Based on the order of the contour area from largest to smallest, one of the predicted boxes is taken out sequentially from the predicted box area sequence as the box to be classified. Calculate the difference between the coordinate information corresponding to the box to be classified and the coordinate information corresponding to the remaining predicted boxes in the predicted box area sequence; The prediction frames whose differences are below a preset error range are assigned to the associated prediction groups corresponding to the same photovoltaic panel.

6. The image recognition result filtering method as described in claim 5, characterized in that, The coordinate information includes: contour coordinates and center point coordinates. The step of calculating the difference between the coordinate information corresponding to the box to be classified and the coordinate information corresponding to the remaining predicted boxes in the predicted box area sequence includes: The area overlap rate is calculated based on the first contour coordinates corresponding to the box to be classified and the second contour coordinates corresponding to the remaining predicted boxes in the predicted box area sequence. The center point distance is calculated based on the first center point coordinates corresponding to the box to be classified and the second center point coordinates corresponding to the remaining predicted boxes in the predicted box area sequence. The difference is obtained based on the area overlap rate or the center point distance.

7. The image recognition result filtering method as described in claim 5, characterized in that, The step of assigning prediction frames with differences below a preset error range to the associated prediction groups corresponding to the same photovoltaic panel includes: Remove predicted boxes whose difference is below a preset error range from the predicted box area sequence; The difference comparison is performed sequentially until each prediction box in the prediction box area sequence is assigned to the corresponding associated prediction group.

8. An image recognition result filtering device, characterized in that, The device includes: An image acquisition module is used to acquire a cropped sub-image, wherein the cropped sub-image contains at least one photovoltaic panel; The coordinate recognition module is used to input the cropped sub-image into the photovoltaic panel inference model to obtain the coordinate information of at least one prediction box. The prediction box is a rectangular box that selects each outline of the photovoltaic panel and is recognized and output by the photovoltaic panel inference model. An area sorting module is used to calculate the contour area of ​​the prediction box corresponding to each of the coordinate information, and sort the contour areas to obtain a prediction box area sequence. The prediction grouping module is used to traverse the coordinate information corresponding to each prediction box in the prediction box area sequence to obtain the associated prediction group of each photovoltaic panel. Each associated prediction group contains all the prediction boxes of the corresponding photovoltaic panel. The result recognition module is used to select the prediction box with the largest outline area in each of the associated prediction groups as the screening result of the target photovoltaic panel.

9. An image recognition result filtering device, characterized in that, The image recognition result filtering device includes: a memory, a processor, and an image recognition result filtering processing program stored in the memory and executable on the processor. When the image recognition result filtering processing program is executed by the processor, it implements the steps of the image recognition result filtering method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the image recognition result filtering method as described in any one of claims 1 to 7.

Images