Flat uniaxial purline and detection method of mounting hole thereof

Abstract

The invention relates to a method for detecting a flat uniaxial purline and a mounting hole thereof. The method comprises the following steps: acquiring image data of the purline and the mounting hole thereof under different working conditions; carrying out Df ine model training on the acquired image data; and predicting the test set by using the trained weight to obtain a prediction result of the purline and the mounting hole. According to the method, the image data of the purline and the mounting hole thereof under different working conditions are collected, and the target detection algorithm based on the Df ine model is adopted, so that high-precision detection of the purline and the mounting hole is realized, and the problem of low detection precision of the flat uniaxial purline and the mounting hole thereof is solved.

Description

Technical Field

[0001] This invention relates to the field of automated installation technology for photovoltaic modules, specifically to a method for detecting flat single-axis purlins and their mounting holes. Background Technology

[0002] In the automated installation process of photovoltaic modules, high-precision inspection of the purlins and their mounting holes at the installation location is required. However, factors such as sunlight, dirt, and rust can affect the accuracy of the inspection. Summary of the Invention

[0003] The purpose of this invention is to overcome the problems existing in the prior art and provide a method for detecting flat single-shaft purlins and their mounting holes, aiming to solve the problem of low detection accuracy of flat single-shaft purlins and their mounting holes.

[0004] To achieve the above-mentioned technical objectives and effects, the present invention is implemented through the following technical solution:

[0005] A method for detecting a flat single-shaft purlin and its mounting hole, the method comprising the following steps:

[0006] Step S1: Collect image data of purlins and their mounting holes under different illumination levels;

[0007] Step S2: Train the acquired image data based on the Dfine model, and analyze the evaluation parameters of the training results;

[0008] Step S3: Use the trained weights to predict the test set to obtain the prediction results of the purlins and mounting holes, and analyze the prediction results.

[0009] Furthermore, in step S1, on the one hand, an illuminance meter is used to measure the intensity of light under different weather conditions, and a camera is used to collect images of the purlins and their mounting holes under different light intensities, angles, and heights. On the other hand, images of rusted or dirty purlins are collected at the construction site.

[0010] Furthermore, in step S2, the collected image data is augmented to expand the dataset, and then the purlins and mounting holes are labeled. After labeling, the dataset is divided into training set, test set, and validation set for training based on the Dfine model. At the same time, the relationship between the purlins and mounting holes is limited to an inclusion relationship during model training, and the detection box of the mounting hole must be within the detection box of the purlin to reduce false detections and false negatives of the mounting holes and improve the detection accuracy.

[0011] The beneficial effects of this invention are:

[0012] This invention achieves high-precision detection of purlins and their mounting holes by collecting image data of purlins and their mounting holes under different working conditions and using a target detection algorithm based on the Dfine model, thus solving the problem of low detection accuracy of flat single-axis purlins and their mounting holes. Attached Figure Description

[0013] Figure 1 This is a schematic flowchart of the method of the present invention;

[0014] Figure 2 This is a schematic diagram illustrating the inclusion relationship between purlins and mounting holes during training in the model defined by this invention.

[0015] Figure 3 This refers to image data under different illumination levels and overhead viewing conditions according to the present invention;

[0016] Figure 4 Images enhanced by data according to the present invention;

[0017] Figure 5 Images acquired and labeled for this invention;

[0018] Figure 6 The image shows the detection when the mounting hole is not obstructed.

[0019] Figure 7 This is a detection image when the mounting hole is partially obscured. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0021] like Figure 1 As shown, a method for inspecting a flat single-shaft purlin and its mounting hole includes the following steps:

[0022] Step S1: Measure the outdoor light intensity (lux) using a lux meter, with the unit being lx. 1–1000 lx is classified as extremely low illuminance, 1000–3000 lx as low illuminance, 3000–8000 lx as low-to-medium illuminance, 8000–15000 lx as medium illuminance, 15000–25000 lx as medium-to-high illuminance, 25000–80000 lx as high illuminance, and 80000–160000 lx as extremely high illuminance. The data acquisition device is an Orbbec Gemini 335L depth camera with a resolution of 640*480, a shooting distance of 0.5m–1m, and a shooting angle of looking upwards. Figure 3 The image shown is an image of the purlin and its mounting holes under different illumination levels; the augmented data is... Figure 4 As shown.

[0023] Step S2: As Figure 5As shown, the image data is labeled and the dataset is divided. The collected image data is trained based on the Dfine model, and the evaluation parameters of the training are analyzed.

[0024] Step S3: Use the trained weights to predict the test set to obtain the prediction results of the purlins and mounting holes, and analyze the prediction results.

[0025] In step S1, on the one hand, a lux meter is used to measure the intensity of light under different weather conditions, and a camera is used to collect images of the purlins and their mounting holes under different light intensities, angles, and heights. On the other hand, images of rusty or dirty purlins are collected at the construction site.

[0026] In step S2, as Figure 3 and Figure 4 As shown, image enhancement is performed on the collected image data to expand the dataset. Then, purlins and mounting holes are labeled. After labeling, the dataset is divided into training, testing, and validation sets for training the deep learning model. During model training, the relationship between purlins and mounting holes is limited to an inclusion relationship, as shown below. Figure 2 As shown, the detection box for the mounting hole must be within the detection box for the purlin to reduce false positives and false negatives, thus improving detection accuracy. In this embodiment, the following table shows the evaluation metrics after training, where P represents the proportion of correctly predicted boxes detected in all predicted boxes; P(hole) represents the accuracy of the mounting hole prediction; R represents the proportion of correctly predicted boxes in all labeled boxes; AP represents the average precision value for a single class; mAP represents the average AP of all classes of purlins and mounting holes in the dataset. The higher the value of mAP, the better the detection performance of the model; mAP50 indicates that the IOU threshold is set to 0.5, where IOU is the degree of overlap between the predicted bounding box and the ground truth bounding box; TP represents the number of correctly matched predicted boxes in all labeled boxes; FP represents the number of incorrectly predicted boxes; FN represents the number of falsely detected labeled boxes; and n represents the class. The calculation formula is as follows:

[0027] P = TP / (TP + FP)

[0028] R = TP / (TP + FN)

[0029]

[0030] Evaluation indicators P P(hole) mAP50 R Deep learning models 0.978 0.963 0.981 0.974

[0031] The table above shows that the accuracy of the Dfine model in predicting purlins and mounting holes is 97.8%, the accuracy in predicting mounting holes in a single category is 96.3%, the average precision is 98.1%, and the recall rate is 97.4%.

[0032] as follows Figure 6and Figure 7 The results shown are the predictions after training the weights. The analysis shows that the position of the mounting hole can be identified both when the mounting hole is not obstructed and when it is partially obstructed.

[0033] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting a flat single-shaft purlin and its mounting holes, characterized in that, The method includes the following steps: Step S1: Collect image data of purlins and their mounting holes under different illumination levels; Step S2: Train the acquired image data based on the Dfine model, and analyze the evaluation parameters of the training results; Step S3: Use the trained weights to predict the test set to obtain the prediction results of the purlins and mounting holes, and analyze the prediction results.

2. The method for detecting the flat single-shaft purlin and its mounting hole according to claim 1, characterized in that, In step S1, on the one hand, a lux meter is used to measure the intensity of light under different weather conditions, and a camera is used to collect images of the purlins and their mounting holes under different light intensities, angles, and heights. On the other hand, images of rusty or dirty purlins are collected at the construction site.

3. The method for detecting the flat single-shaft purlin and its mounting hole according to claim 1, characterized in that, In step S2, the collected image data is augmented to expand the dataset. Then, the purlins and mounting holes are labeled. After labeling, the dataset is divided into training set, test set, and validation set for training based on the Dfine model. At the same time, the relationship between the purlins and mounting holes is limited to an inclusion relationship during model training. The detection box of the mounting hole must be within the detection box of the purlin to reduce false detections and false negatives of the mounting holes and improve the detection accuracy.

Images