Coal mine underground drill site rod retreating counting method

Abstract

The invention discloses an underground coal mine drill site rod retreating counting method, which is based on a sensing detection technology of a pressure transmitter and an AI video recognition algorithm of YOLOv8, and specifically comprises the following steps of: 1, collecting data; step 2, data annotation; step 3, data enhancement; 4, model training is carried out; 5, outputting the model; step 6, performing model reasoning and obtaining a sensing detection result of the pressure transmitter; and 7, fusing analysis results: fusing an AI analysis result of the YOLOv8 and a sensing technology analysis result together to form a fusion analysis strategy, and finally obtaining the number of withdrawn rods of the underground drill site of the coal mine. According to the coal mine underground drill site rod retreating counting method, the intelligent sensing technology and the advanced artificial intelligence algorithm are combined, and the accuracy and reliability of drill rod number recognition are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of the recognition of the number of drill rods withdrawn in a coal mine underground drill site, and in particular to a method for counting the number of drill rods withdrawn in a coal mine underground drill site. Background Art

[0002] In coal mines, the recognition of the number of drill rods withdrawn during the construction of a gas drainage drill site is of great significance in the drilling operation. By recognizing the number of drill rods withdrawn, it can be ensured whether the drilling operation is carried out according to the designed hole depth, and behaviors such as false drilling and false reporting of the footage can be avoided. Therefore, the accurate recognition of the number of drill rods withdrawn is crucial for ensuring the safety of coal mine production. With the development of technology, the combination of intelligent sensing technology and advanced artificial intelligence algorithms can further improve the accuracy and reliability of drill rod number recognition.

[0003] The present invention discloses a method for counting the number of drill rods withdrawn in a coal mine underground gas drainage drill site based on a pressure transmitter and the YOLOv8 algorithm. Since the environment in the drilling operation area is harsh, there may be situations such as occlusion and reflection during the drill rod withdrawal process, which affect the results of AI video analysis. It is considered to combine sensing technology and AI video analysis technology to count the number of drill rods withdrawn. The pressure transmitter is used to detect the pressure received by the gripper that holds the drill rod on the drilling rig, and whether multiple drill rod withdrawal actions are completed is analyzed by combining the pressure changes during the drill rod withdrawal process. The YOLOv8 algorithm is an object detection algorithm that can be used to detect drilling rigs, drill rods, and personnel, and AI drill rod withdrawal counting is performed through rules such as the change in the length of the detected drill rods. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art.

[0005] For this reason, the present invention proposes a method for counting the number of drill rods withdrawn in a coal mine underground drill site, which combines intelligent sensing technology and advanced artificial intelligence algorithms to improve the accuracy and reliability of drill rod number recognition.

[0006] A method for counting the number of drill rods withdrawn in a coal mine underground drill site according to an embodiment of the present invention, based on the sensing detection technology of a pressure transmitter and the AI video recognition algorithm of YOLOv8, specifically includes the following steps:

[0007] Step 1, data collection: After using a camera to shoot a drill rod withdrawal video, extract frames to collect pictures;

[0008] Step 2, data annotation: Annotate the types and target frames of the drilling rig, drill rods, and people, and perform annotation;

[0009] Step 3, data augmentation: Generate new training samples by performing various transformations and processing on the existing data, so as to increase the diversity and quantity of the data set;

[0010] Step 4, model training;

[0011] Step 5, Output the model;

[0012] Step 6, Model inference and sensing detection results of the pressure transmitter:

[0013] The specific steps of the model inference are as follows:

[0014] Step 61-1, Load the trained model;

[0015] Step 62-1, Load real-time images;

[0016] Step 63-1, Model inference;

[0017] Step 64-1, Post-process the target bounding boxes;

[0018] Step 65-1, AI analysis results of YOLOv8;

[0019] The specific steps of the sensing detection results of the pressure transmitter are as follows:

[0020] Step 61-2, Access the pressure data;

[0021] Step 62-2, Obtain and analyze the change of pressure data near the abnormal time point;

[0022] Step 63-2, Generate sensing technology analysis results;

[0023] Step 7, Fusion analysis results: Integrate the AI analysis results of YOLOv8 and the sensing technology analysis results to form a fusion analysis strategy, and finally obtain the number of drill rod withdrawals in the coal mine underground drill site.

[0024] According to an embodiment of the present invention, in the said Step 4, model training includes training data, loss function, learning algorithm, error, and optimizer;

[0025] The training data is the dataset for training the model;

[0026] The loss function is used to evaluate the performance of the model;

[0027] The learning algorithm is used to update the model parameters to minimize the loss function;

[0028] The error is used to describe the overall prediction accuracy and generalization ability of the model;

[0029] The optimizer is used to adjust the weights and biases of the neural network.

[0030] According to an embodiment of the present invention, the said Step 4 specifically includes the following steps:

[0031] Step 41, Obtain the output by forward propagating data points through the network;

[0032] Step 42: Calculate the total error through the loss function;

[0033] Step 43: Use the backpropagation algorithm to calculate the gradients of the loss function with respect to each weight and bias;

[0034] Step 44: Use the gradient descent algorithm to update the weights and biases of each layer;

[0035] Step 45: Repeat Steps 41 to 44 to minimize the total error.

[0036] According to an embodiment of the present invention, in the 7th step, during the drill-back process, the AI analyzes the number of drill rods withdrawn, filters the drill rods using the drill rig target, and accesses the data of the pressure transmitter to assist in verifying the number of drill rods withdrawn.

[0037] According to an embodiment of the present invention, in the 7th step, the processing flow of the fusion analysis strategy is as follows:

[0038] Step 71: Obtain the detection information of the drill rig, drill rods, and people;

[0039] Step 72: Filter by detecting the connection point characteristics between the drill rod and the coal washing pipeline;

[0040] Step 73: When the connection point between the drill rod and the coal washing pipeline is detected, do not count;

[0041] Step 74: When the area where the drill rod is unloaded is blocked or has strong reflection, resulting in unclear or indistinguishable changes in the length of the drill rod, record that moment and generate an evidence screenshot;

[0042] Step 75: Analyze the change process of the data of the accessed pressure transmitter at the above-recorded moment to update the existing drill rod counting result.

[0043] According to an embodiment of the present invention, in the 3rd step, data augmentation includes random rotation, random scaling, random cropping, horizontal flipping, brightness, contrast, and color adjustment, adding noise, translation, and transformation of angles.

[0044] According to an embodiment of the present invention, in the 65-1st step, the AI analysis results of YOLOv8 include the drill rod withdrawal counting result and the abnormal time points when it is detected that there is strong occlusion or reflection and the connection to the coal washing pipeline.

[0045] According to an embodiment of the present invention, the YOLOv8 network finally outputs the types of detected targets, the target rectangular frames, and the drill rod withdrawal counting information. Among them, the types are discrete and calculated using the classification loss function, and the target box information is continuous and calculated using the CIoU and DFLoss loss functions.

[0046] According to an embodiment of the present invention, the calculation formula of the classification loss function L is as follows:

[0047]

[0048] Among them, the meanings represented by each symbol in the above formula are:

[0049] L represents the classification loss function;

[0050] N represents the number of samples;

[0051] L I represents the loss of the i-th sample;

[0052] y I represents the true value of the i-th sample;

[0053] p i represents the predicted value of the i-th sample.

[0054] According to an embodiment of the present invention, the calculation formula of the CIoU is as follows:

[0055]

[0056] Among them, the meanings represented by each symbol in the above formula are:

[0057] CIoU represents the loss function for object detection, mainly used to evaluate the similarity between the predicted bounding box and the true bounding box;

[0058] IoU represents the intersection over union of the predicted bounding box and the true bounding box; specifically, in the field of object detection, IoU represents the overlap rate between the predicted bounding box and the true bounding box, that is, the intersection over union;

[0059] ρ represents the Euclidean distance between the centers of the predicted bounding box and the true bounding box;

[0060] b represents the center point of the predicted bounding box;

[0061] b gt represents the center point of the true bounding box;

[0062] c represents the diagonal length of the smallest circumscribed rectangle that can simultaneously contain the predicted bounding box and the true bounding box;

[0063] α represents the weight function;

[0064] v represents a measure of the consistency of the aspect ratio;

[0065] The calculation formula of the DFLoss is as follows:

[0066] DFLoss(Si ,S i+1 )=-((y i+1 -y)log(S i )+(y - y i )log(S i+1 ))

[0067] Among them, the meanings of the symbols in the above formula are as follows:

[0068] DFLoss represents the regression of the predicted bounding box in a probabilistic manner, aiming to solve the class imbalance problem and improve the detection accuracy;

[0069] S i represents the probability distribution of the difference between the left side of the predicted bounding box and the true value;

[0070] S i+1 represents the probability distribution of the difference between the right side of the predicted bounding box and the true value;

[0071] y represents the true value of the target box of the i-th sample;

[0072] y i represents the ceiling of the number fused by the prediction;

[0073] y i+1 represents the floor of the number fused by the prediction.

[0074] The beneficial effect of the present invention is that based on the pressure transmitter and the YOLOv8 algorithm, the pressure transmitter is used to detect the pressure received by the gripper that holds the drill pipe on the drill rig. By analyzing the pressure change during the drill pipe withdrawal process, it is determined whether multiple drill pipe withdrawal actions are completed. The YOLOv8 algorithm is an object detection algorithm that can be used to detect the drill rig, drill pipe, and personnel. By detecting rules such as the change in the length of the drill pipe, AI drill pipe withdrawal counting is performed. Finally, the sensing technology and AI video analysis technology are combined to count the drill pipe withdrawal, greatly improving the accuracy and reliability of drill pipe number recognition.

[0075] Other features and advantages of the present invention will be described in the following specification, and part of them will become obvious from the specification or be understood by implementing the present invention. The objectives and other advantages of the present invention are achieved and obtained by the structures specifically pointed out in the specification and the drawings.

[0076] To make the above objectives, features, and advantages of the present invention more obvious and understandable, the following specific preferred embodiments are given and described in detail in conjunction with the accompanying drawings. Description of the Drawings

[0077] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in this application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0078] Figure 1 is a schematic diagram of the backreaming scenario;

[0079] Figure 2 is a technical architecture diagram of the drill pipe counting method;

[0080] Figure 3 is a detailed structure diagram of the YOLOv8 network;

[0081] Figure 4 is a flowchart of the model training of the present invention;

[0082] Figure 5 is a flowchart of the model inference of the present invention;

[0083] Figure 6 is a flowchart of obtaining the sensing detection result of the present invention;

[0084] Figure 7 is a flowchart of the fusion analysis result of the present invention;

[0085] Figure 8 is a schematic diagram of the backreaming process Figure 1 ;

[0086] Figure 9 is a schematic diagram of the backreaming process Figure 2 ;

[0087] Figure 10 is a schematic diagram of the backreaming process Figure 3 ;

[0088] Figure 11 is a labeled effect diagram;

[0089] Figure 12 is a schematic diagram of the deep learning network structure;

[0090] Figure 13 is an effect diagram of the present invention. Specific implementation manners

[0091] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0092] Next, a method for counting the retracted drill rods in a coal mine underground drill site according to an embodiment of the present invention will be specifically described with reference to the accompanying drawings.

[0093] See Figure 1-13 , a method for counting the retracted drill rods in a coal mine underground drill site according to the present invention is based on the sensing detection technology of a pressure transmitter and the AI video recognition algorithm of YOLOv8, and specifically includes the following steps:

[0094] Step 1, data collection: After using a camera to shoot the video of the retracted drill rod, extract frames to collect pictures, and collect the process of retracting the drill as Figure 8-10 shown; among them, Figure 8 shows that the gripper is loosening the drill rod and the worker is waiting to remove the drill rod; Figure 9 shows the action of the worker removing the drill rod; Figure 10 shows the worker taking down the drill rod and leaning it on the ground.

[0095] Step 2, data annotation: Annotate the types and target frames of the drill rig, drill rod, and person, and perform annotation; it should be noted that during the data annotation process, the drill rig, drill rod, and person are divided into three categories. Annotate the annotation information of the three target categories of the drill rig, drill rod, and person. The annotation information includes the target category and the target rectangle frame, and the annotation effect is as Figure 11 shown. Among them, Figure 11 the purple frame, red frame, and white frame in

[0096] are all target frames. The purple frame corresponds to the drill rig, the red frame corresponds to the drill rod, and the white frame corresponds to the person.

[0097] Data Augmentation is a technique that generates new training samples by performing various transformations and processes on existing data, thereby increasing the diversity and quantity of the dataset. These transformations can include geometric transformations, color transformations, noise addition, etc., enabling the model to encounter a greater variety of data during training, thus enhancing the model's generalization ability and robustness. Data Augmentation can be applied to various types of data, including images, text, audio, etc. For image data, common data augmentation operations include random rotation, random scaling, random cropping, horizontal flipping, brightness, contrast, and color adjustment, noise addition, translation, and transformation of angles.

[0098] Random Rotation: Randomly rotate the image by a certain angle to simulate shooting at different angles.

[0099] Random Scaling: Randomly enlarge or reduce the size of the image to simulate shooting at different distances.

[0100] Random Cropping: Randomly crop a part of the image to simulate shooting with different fields of view.

[0101] Horizontal Flipping: Randomly flip the image horizontally to simulate mirror shooting.

[0102] Brightness, Contrast, and Color Adjustment: By adjusting the brightness, contrast, and color of the image, make the model more robust to illumination and color changes.

[0103] Noise Addition: Add random noise to the image to increase the model's robustness to noise.

[0104] Translation: Randomly translate the position of the image to simulate different shooting angles.

[0105] Transformation of Angles: Perform projective transformation, perspective transformation, etc. on the image to increase the diversity of image transformations.

[0106] Data Augmentation is a commonly used technique in machine learning, especially in deep learning. By applying a series of transformations to the original data, such as rotation, scaling, cropping, color adjustment, etc., the diversity of the data is increased, thereby enhancing the model's generalization ability.

[0107] Step 4, Model Training;

[0108] Model training includes training data, loss function, learning algorithm, error, and optimizer.

[0109] Train Data: The dataset used to train the model, enabling the model to learn the patterns in the data.

[0110] Loss Function: A function that measures the difference between the predicted result and the true result, also known as the error function. It evaluates the performance of the model by calculating the degree of inconsistency between the predicted value and the true value of the model.

[0111] Learning Algorithm: An algorithm used to update the model parameters to minimize the loss function. In deep learning, the commonly used learning algorithm is Gradient Descent and its variants. By calculating the gradient of the loss function with respect to the model parameters and updating the parameters in the opposite direction of the gradient, the value of the loss function is gradually reduced.

[0112] Error: Describes the overall prediction accuracy and generalization ability of the model. By adjusting the Bias and Variance, the Error can be reduced and the prediction ability of the model can be improved.

[0113] Optimizer: Responsible for adjusting the weights and biases of the neural network to minimize the loss function, thereby improving the accuracy and performance of the model.

[0114] The specific model training includes the following steps:

[0115] Step 41: Obtain the output by forward propagating the data points through the network;

[0116] Step 42: Calculate the total error through the loss function;

[0117] Step 43: Use the backpropagation algorithm to calculate the gradient of the loss function with respect to each weight and bias;

[0118] Step 44: Use the gradient descent algorithm to update the weights and biases of each layer;

[0119] Step 45: Repeat Steps 41 to 44 to minimize the total error.

[0120] The YOLOv8 network finally outputs the types of detected objects, the target rectangle boxes, and the information of the retracting rod count. Among them, the types are discrete and calculated using the classification loss function. The target box information is continuous and the loss function is calculated using CIoU + DFLoss.

[0121] Classification Loss: Use BCE (Binary Cross Entropy) as the classification loss. For each category, judge "whether it is this category" and output the confidence.

[0122]

[0123] Among them, the meanings represented by each symbol in the above formula are:

[0124] L represents the classification loss function;

[0125] N represents the number of samples;

[0126] L i represents the loss of the i-th sample;

[0127] y i represents the true value of the i-th sample;

[0128] p i represents the predicted value of the i-th sample.

[0129] The calculation formula of CIoU is as follows:

[0130]

[0131] Among them, the meanings represented by each symbol in the above formula are:

[0132] CIoU represents the loss function for object detection, mainly used to evaluate the similarity between the predicted bounding box and the true bounding box;

[0133] IoU represents the intersection over union of the predicted bounding box and the true bounding box; specifically, in the field of object detection, IoU represents the overlap rate between the predicted bounding box and the true bounding box, that is, the intersection over union;

[0134] ρ represents the Euclidean distance between the centers of the predicted bounding box and the true bounding box;

[0135] b represents the center point of the predicted bounding box;

[0136] b gt represents the center point of the true bounding box;

[0137] c represents the diagonal length of the smallest circumscribed rectangle that can simultaneously contain the predicted bounding box and the true bounding box;

[0138] α represents the weight function;

[0139] v represents the consistency used to measure the aspect ratio.

[0140]

[0141] Among them, the meanings represented by each symbol in the above formula are:

[0142] Intersection represents the area of the intersection part of the two bounding boxes of the true bounding box and the predicted bounding box;

[0143] box1 represents the true bounding box;

[0144] box2 represents the predicted bounding box;

[0145]

[0146] Among them, the meanings represented by each symbol in the above formula are as follows:

[0147] w gt represents the true border width;

[0148] h gt represents the true border height;

[0149] w represents the predicted border width;

[0150] h represents the predicted border height.

[0151] DFLoss(S i ,S i+1 )=-((y i+1 -y)log(S i )+(y - y i )log(S i+1 ))

[0152] Among them, the meanings represented by each symbol in the above formula are as follows:

[0153] DFLoss represents the regression of the predicted border in a probabilistic manner, aiming to solve the class imbalance problem and improve the detection accuracy;

[0154] S i represents the probability distribution of the difference between the left side of the predicted border and the true value;

[0155] S i+1 represents the probability distribution of the difference between the right side of the predicted border and the true value;

[0156] y represents the true value of the target box of the i-th sample;

[0157] y i represents the ceiling of the number fused by the prediction;

[0158] y i+1 represents the floor of the number fused by the prediction.

[0159] Step 5, output the model;

[0160] Step 6, model inference and the sensing detection result of the pressure transmitter:

[0161] The specific steps of the model inference are as follows:

[0162] Step 61-1, load the training model: Pre-load the pre-trained model of the drilling site;

[0163] Step 62-1, load the real-time picture: Use the camera to collect pictures, where the collected pictures are the video images of the on-site drilling operation;

[0164] Step 63-1, Model Inference: By detecting the drill rig, drill pipe, and people, and identifying changes in the length of the drill pipe, the rod withdrawal count during the rod withdrawal process is achieved. At the same time, it is detected whether there is a situation where the drill pipe is blocked, there is reflection, or the connection of the coal washing pipeline affects the inability to identify the drill pipe. If so, it is recorded and saved to the local database to generate evidence pictures.

[0165] Step 64-1, Post-processing of Target Boxes: Further process and optimize the detected target boxes, including non-maximum suppression, bounding box regression and adjustment, multi-object tracking, etc.;

[0166] Step 65-1, AI Analysis Results of YOLOv8: Video screenshots showing the influence of blocked drill pipes, reflection, and connection of the coal washing pipeline are analyzed from the video images, including the rod withdrawal count results and the abnormal time points when the drill pipe is blocked, has strong reflection, or is connected to the coal washing pipeline, etc.;

[0167] The specific steps of the sensing detection results of the pressure transmitter are as follows:

[0168] Step 61-2, Pressure Data Access: Access the real-time data of the gripper pressure collected by the pressure transmitter through the RS485 transmission protocol;

[0169] Step 62-2, Pressure Data Analysis: Obtain and analyze the change of pressure data near the abnormal time point; specifically, analyze the change of the gripper pressure collected according to the abnormal time point of the video screenshots showing the influence of blocked drill pipes, reflection, and connection of the coal washing pipeline;

[0170] Step 63-2, Generate Sensing Technology Analysis Results: According to the change of the gripper pressure near the abnormal time point, analyze whether there is a rod withdrawal; if so, the rod withdrawal count is incremented by 1; otherwise, the rod withdrawal number remains unchanged.

[0171] Step 7, Fusion Analysis Results: Integrate the AI analysis results of YOLOv8 and the sensing technology analysis results to form a fusion analysis strategy, and finally obtain the rod withdrawal number in the underground coal mine drill site;

[0172] The rod withdrawal number in the AI analysis of the rod withdrawal process can be obtained by rules such as detecting changes in the length of the drill pipe. However, due to factors such as many objects with similar drill pipe characteristics, reflection, and occlusion in the drill site, in order to prevent misdetection, the drill pipe is filtered using the drill rig target and the data of the pressure transmitter is accessed to assist in verifying the rod withdrawal number. The processing flow of the fusion analysis strategy is as follows:

[0173] Step 71, Obtain the detection information of the drill rig, drill pipe, and people;

[0174] Step 72: Filter by detecting the connection point characteristics between the drill pipe and the coal washing pipeline;

[0175] Step 73: When the connection point between the drill pipe and the coal washing pipeline is detected, do not count; specifically, by detecting the connection point between the drill pipe and the coal washing pipeline, determine whether the coal washing pipeline has been connected. If the coal washing pipeline has been connected, it is considered that the rod withdrawal action has not been performed;

[0176] Step 74: When the area where the drill pipe is unloaded is blocked or has strong reflection, resulting in an unclear or indistinguishable change in the length of the drill pipe, record this moment and generate an evidence screenshot;

[0177] Step 75: Analyze the change process of the data of the connected pressure transmitter near the above recording moment, and update the existing drill pipe counting result.

[0178] See Figure 2 , the technical architecture diagram of the drill pipe counting method involves the AI video recognition algorithm based on YOLOv8 and the sensing detection technology based on the pressure transmitter. For the pressure transmitter, the rod withdrawal action is driven by controlling the oil pressure change, and the completed rod withdrawal action can be analyzed according to the detected pressure change of the gripper. For the YOLOv8 network structure: Backbone is responsible for feature extraction, using a series of convolutional and deconvolutional layers, and at the same time using residual connections and bottleneck structures to reduce the size of the network and improve performance; this part uses the C2f module as the basic building unit. Compared with the C3 module of YOLOv5, the C2f module has fewer parameters and better feature extraction ability. Neck is responsible for multi-scale feature fusion, by fusing the feature maps from different stages of Backbone to enhance the feature representation ability. Head is responsible for the final object detection and classification tasks, including a detection head, a classification head, and a counting recognition head; the detection head contains a series of convolutional and deconvolutional layers for generating detection results; the classification head uses global average pooling to classify each feature map; the counting recognition head is used to identify the number of rod withdrawals.

[0179] See Figure 12 , Input Layer represents the input layer, Hidden Layer represents the hidden layer, and Output Layer represents the output layer.

[0180] See Figure 13 , Time represents the detection pressure recording time point, and Pressure represents the gripper pressure data corresponding to the time point.

[0181] A method for counting the number of drill rods withdrawn in a coal mine underground drill site according to the present invention not only solves the problem that the accuracy of the traditional single video analysis algorithm fluctuates greatly due to factors such as occlusion, reflection, and similar objects, but also solves the problem of identifying the number of drill rods withdrawn in the drill site. It is based on a pressure transmitter and the YOLOv8 algorithm. The pressure transmitter is used to detect the pressure received by the gripper that holds the drill rod on the drill rig. By analyzing the pressure change during the process of withdrawing the drill, it is determined whether multiple drill rod withdrawal actions are completed. The YOLOv8 algorithm is an object detection algorithm that can be used to detect drill rigs, drill rods, and personnel. Through rules such as the detected change in the length of the drill rod, AI drill rod withdrawal counting is performed. Finally, by combining sensing technology and AI video analysis technology for drill rod withdrawal counting, the accuracy and reliability of drill rod number identification are greatly improved.

[0182] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A method for counting rod withdrawal in a drilling field in an underground coal mine, characterized in that: The sensor detection technology based on the pressure transmitter and the AI ​​video recognition algorithm of YOLOv8 specifically includes the following steps: Step 1, data collection: Use a camera to shoot a video of the back-pole and then extract frames to collect images; Step 2: Data labeling: label the drilling rig, drill rod, type of person and target frame, and label them; Step 3, data enhancement: Generate new training samples by performing various transformations and processing on existing data, thereby increasing the diversity and quantity of the data set; Step 4: Model training; Step 5: Output model; Step 6: Model reasoning and pressure transmitter sensing results: The specific steps of the model reasoning are as follows: Step 61-1, training model loading; Step 62-1, real-time image loading; Step 63-1, model reasoning; Step 64-1, target frame post-processing; Step 65-1, YOLOv8 AI analysis results; The specific steps of sensing the detection result of the pressure transmitter are as follows: Step 61-2: Pressure data access; Step 62-2, obtaining and analyzing the pressure data changes near the abnormal time point; Step 63-2, generating the sensor technology analysis results; Step 7, Fusion Analysis Results: Fusion the YOLOv8 AI analysis results and the sensor technology analysis results to form a fusion analysis strategy, and ultimately obtain the number of rod withdrawals in the underground coal mine drilling field.

2. A method for counting rod withdrawal in a coal mine drilling field according to claim 1, characterized in that: In the fourth step, model training includes training data, loss function, learning algorithm, error and optimizer; The training data is a data set used to train the model; The loss function is used to evaluate the performance of the model; The learning algorithm is used to update the model parameters to minimize the loss function; The error is used to describe the overall prediction accuracy and generalization ability of the model; The optimizer is used to adjust the weights and biases of the neural network.

3. A method for counting rod withdrawal in a coal mine drilling field according to claim 1 or 2, characterized in that: The fourth step specifically includes the following steps: Step 41, forward propagating the data points through the network to obtain the output; Step 42, calculating the total error through the loss function; Step 43: Use the back-propagation algorithm to calculate the gradient of the loss function with respect to each weight and bias. Step 44: Update the weights and biases of each layer using the gradient descent algorithm. Step 45: Repeat steps 41 to 44 to minimize the total error.

4. A method for counting rod withdrawal in a drilling field in an underground coal mine according to claim 1, characterized in that: In the seventh step, the AI ​​analyzes the number of rods withdrawn during the drill withdrawal process, uses the drilling rig target to filter the drill rods, and accesses the pressure transmitter data to assist in verifying the number of rods withdrawn.

5. A method for counting rod withdrawal in a coal mine drilling field according to claim 1 or 4, characterized in that: In the seventh step, the processing flow of the fusion analysis strategy is as follows: Step 71: Obtain detection information of the drilling rig, the drill rod, and the person; Step 72: filtering by detecting the connection point characteristics between the drill pipe and the coal flushing pipeline; Step 73: when the connection point between the drill pipe and the coal flushing pipeline is detected, no counting is performed; Step 74: When the drill rod unloading area is blocked or has strong reflections, so that the change in the length of the drill rod is not obvious or cannot be distinguished, record the moment and generate an evidence screenshot; Step 75: By analyzing the change process of the connected pressure transmitter data at the above recording time, the existing drill rod counting result is updated.

6. A method for counting rod withdrawal in a coal mine drilling field according to claim 1, characterized in that: In the third step, data enhancement includes random rotation, random scaling, random cropping, horizontal flipping, brightness, contrast and color adjustment, noise addition, translation and angle change.

7. A method for counting rod withdrawal in a coal mine drilling field according to claim 1, characterized in that: In the step 65-1, the AI ​​analysis results of YOLOv8 include the rod retreat counting results and the abnormal time points when obstruction or strong reflection and connection to the coal flushing pipeline are detected.

8. A method for counting rod withdrawal in a coal mine drilling field according to claim 1, characterized in that: The YOLOv8 network finally outputs the type of detected target, target rectangular frame and back-off count information. The categories are discrete and are calculated using the classification loss function. The target box information is continuous and the loss function is calculated using CIoU and DFLoss.

9. A method for counting rod withdrawal in a drilling field in an underground coal mine according to claim 8, characterized in that: The calculation formula of the classification loss function L is as follows: The meanings of the symbols in the above formula are: L represents the classification loss function; N represents the number of samples; L i represents the i-th sample loss; y i Represents the true value of the i-th sample; p i Represents the predicted value of the i-th sample.

10. A method for counting rod withdrawal in a drilling field in an underground coal mine according to claim 8, characterized in that: The calculation formula of CIoU is as follows: The meanings of the symbols in the above formula are: CIoU represents the loss function for object detection, which is mainly used to evaluate the similarity between the predicted bounding box and the true bounding box; IoU represents the intersection-over-union ratio of the predicted bounding box and the true bounding box. Specifically, in the field of object detection, IoU represents the overlap ratio between the predicted bounding box and the true bounding box, i.e., the intersection-over-union ratio. ρ represents the Euclidean distance between the center point of the predicted bounding box and the true bounding box; b represents the center point of the predicted bounding box; b gt Indicates the center point of the real bounding box; c represents the diagonal length of the minimum bounding rectangle that can contain both the predicted bounding box and the true bounding box; α represents the weight function; v represents the consistency used to measure the aspect ratio; The calculation formula of the DFLoss is as follows: DFLoss(S i ,S i+1 )=-((y i+1 -y)log(S i )+(y-y i )log(S i+1 )) The meanings of the symbols in the above formula are: DFLoss means regressing the predicted bounding box in a probabilistic way, aiming to solve the class imbalance problem and improve detection accuracy; S i Represents the probability distribution of the difference between the left side of the predicted bounding box and the true value; S i+1 Represents the probability distribution of the difference between the right side of the predicted bounding box and the true value; y represents the true value of the target box of the i-th sample; y i Indicates the upper integer of the predicted fusion number; y i+1 Indicates the number of predicted fusion rounded down.

Images