CN111598042B - Visual statistical method for underground drill rod counting - Google Patents
Visual statistical method for underground drill rod counting Download PDFInfo
- Publication number
- CN111598042B CN111598042B CN202010448311.XA CN202010448311A CN111598042B CN 111598042 B CN111598042 B CN 111598042B CN 202010448311 A CN202010448311 A CN 202010448311A CN 111598042 B CN111598042 B CN 111598042B
- Authority
- CN
- China
- Prior art keywords
- image
- unloading
- video
- data
- convolution
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000000007 visual effect Effects 0.000 title claims abstract description 18
- 238000007619 statistical method Methods 0.000 title claims abstract description 16
- 238000000034 method Methods 0.000 claims abstract description 36
- 238000001514 detection method Methods 0.000 claims abstract description 32
- 230000001965 increasing effect Effects 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims abstract description 14
- 238000001914 filtration Methods 0.000 claims abstract description 13
- 238000012545 processing Methods 0.000 claims abstract description 13
- 230000007306 turnover Effects 0.000 claims abstract description 11
- 238000010606 normalization Methods 0.000 claims abstract description 9
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 230000003044 adaptive effect Effects 0.000 claims abstract description 7
- 230000009467 reduction Effects 0.000 claims abstract description 5
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 18
- 230000008569 process Effects 0.000 claims description 12
- 238000012360 testing method Methods 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 4
- 238000007599 discharging Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000012217 deletion Methods 0.000 claims description 3
- 230000037430 deletion Effects 0.000 claims description 3
- 238000003860 storage Methods 0.000 claims description 3
- 238000009432 framing Methods 0.000 abstract description 2
- 238000005553 drilling Methods 0.000 description 9
- 238000013145 classification model Methods 0.000 description 8
- 239000003245 coal Substances 0.000 description 5
- 230000002708 enhancing effect Effects 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06M—COUNTING MECHANISMS; COUNTING OF OBJECTS NOT OTHERWISE PROVIDED FOR
- G06M11/00—Counting of objects distributed at random, e.g. on a surface
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The invention discloses a visual statistical method for underground drill rod counting, which comprises the following steps: firstly, storing and framing a drill pipe unloading video, and carrying out normalization processing; establishing a binary data set, and dividing the binary data set into corresponding categories of a database according to the content of a single-frame image; the number of data sets is increased by a preprocessing method combining rotation, turnover and brightness enhancement; training an improved adaptive learning rate ResNet-50 network; then detecting the image category of each frame in the video and outputting the confidence percentage; and after all the images pass through a self-adaptive learning rate ResNet-50 model, clearing all the confidence coefficients of the second type non-rod-unloading results in the CSV file, filtering the output confidence coefficient of the video by using an integral method, and finally, calculating the rod-unloading quantity in the video through a falling edge. The method has higher detection precision, can effectively reduce errors caused by model detection, improves the final detection precision, and simultaneously avoids the problem of detection precision reduction caused by shielding.
Description
Technical Field
The invention belongs to the technical field of visual detection, and particularly relates to a visual statistical method for underground drill rod counting.
Background
The gas is used as a large killer which is harmful to the underground production safety of China, and the life safety of underground workers is threatened all the time. In actual production, gas in a coal seam is extracted in a drilling rod drilling mode, quantitative requirements are made on the drilling quantity, and therefore the gas extraction capacity is sufficient. The method has the advantages that the counting of the number of the drill rods is always the problem to be considered when the extraction capacity of the drilled hole is measured, how to automatically count the number of the drill rods is avoided, and the problem that potential hazards brought to production safety due to false alarm and missing alarm of workers are solved at present is avoided, so that the potential safety hazard of gas is reduced, the life safety of underground workers is ensured, and huge property loss of enterprises is reduced.
The existing method for detecting the number of the drill rods mainly adopts artificial counting, and the number of the drill rods is artificially counted by browsing underground drill rod unloading videos in a video monitoring center, so that the method has low automation level and high relative cost; with the development of image processing, the number of image detection drill rods is relatively low in cost and convenient to install, and Chinese patents (application number: 201310132223.9, publication number: CN 104100256B) disclose a coal mine underground drilling depth measuring method based on an image processing technology, wherein the method can identify the front and back actions of the drill rods under the condition of partial shielding, but the video detection cannot expose or shield the marker action for a long time, the monitoring anti-interference performance is poor, and a scheme for further improvement is necessary.
Disclosure of Invention
The invention aims to provide a visual statistical method for underground drill rod counting, which solves the problem of low detection precision caused by low anti-interference performance in the existing detection method.
The technical scheme adopted by the invention is that the visual statistical method for underground drill rod counting is implemented according to the following steps:
collecting a drill pipe unloading video, storing and frame-disassembling the drill pipe unloading video, carrying out normalization processing on images and storing all single-frame images; establishing a two-classification data set, and classifying the two-classification data set into corresponding classes of the database according to the content of the single-frame image;
step two, for unbalanced data quantity and insufficient data quantity of two types in the data set, increasing the data set quantity by a preprocessing method combining rotation, turnover and brightness enhancement;
step three, training an improved adaptive learning rate ResNet-50 network;
inputting a rod unloading video, detecting the type of each frame of image in the video, outputting a confidence percentage, and storing the confidence percentage into a CSV file from an image time sequence;
fifthly, after all the images pass through a self-adaptive learning rate ResNet-50 model, clearing the confidence coefficients of all the second non-rod-unloading results in the CSV file, and filtering the video output confidence coefficient by using an integral method;
counting the falling edges of the filtered 0 and 1 signals, and calculating the rod unloading number in the video through the falling edges;
and (4) counting the falling edges of the signals of 0 and 1, counting the number +1 if the filtering data is changed from 1 to 0, and finally counting the total number of the drill rods.
The present invention is also characterized in that,
in step 1, a binary data set, a first data set S, is established 1 For the unloading drill rod category, a second type data set S 2 And for the non-detachable drill rod category, manually classifying the single-frame images, and setting the proportion of training data to test data to be 8 in a classification data set: 2, classifying the image data volume in the data set to be not less than 2000; during normalization, the original video split frames are normalized in batches into images with the size of 224 × 224; classifying the normalized image into a discharging drill rod category S 1 And non-removable drill pipe category S 2 The image at the non-drill-rod-unloading time is far greater than the image at the drill rod-unloading time, the image at the non-drill-rod-unloading time needs to be deleted randomly, and the quantity ratio of the images in the two types before the deletion is S 1 :S 2 Down to S =1 1 :S 2 And 2, reducing data set unbalance and ensuring that the total data amount is not less than 2000 sheets.
In step 2, the data S of the drill rod being unloaded 1 Data sets are increased through preprocessing, wherein 10% of drill rod unloading data are subjected to rotation operation, 10% of drill rod unloading data are subjected to turnover operation, 20% of drill rod unloading data are subjected to brightness enhancement firstly and rotate, and 20% of drill rod unloading data are subjected to brightness enhancement firstly and rotateFirstly, brightness enhancement and then overturning are carried out on data, and firstly, rotation and then overturning are carried out on 20% of drill rod unloading data; the brightness enhancement adopts Gamma variation, the parameter of the Gamma conversion power is set to be 0.5, and the added parameter is 10.
The third step is specifically as follows:
step 301, extracting a data set and inputting the data set into a ResNet-50 network, wherein the ResNet-50 network structure is composed of a convolution 1, a convolution 2, a convolution 3, a convolution 4, a convolution 5, an average pooling, full connection and a classifier;
a ResNet-50 network structure, wherein the step length of convolution 1 is 2, and the size of convolution kernel is 7 × 7; convolution 2 consists of 3 residual blocks; convolution 3 consists of 4 residual blocks; convolution 4 consists of 6 residual blocks; convolution 5 consists of 3 residual blocks; the activation function of the average pooling is softmax;
the residual block is composed of three layers and is output to the next layer of residual block after 1 × 1 convolution, 3 × 3 convolution and 1 × 1 convolution; the activation function is Relu;
step 302, introducing Logistic empirical formula on the basis of learning rate attenuation
Satisfies the equation-2 a ≈ nb, yields ≈ er>
A formula, wherein n is the total number of learning rounds; the Logistic empirical formula satisfies the equation-2 a ≈ nb, and the value of a is ≈>
n＝100-120，b＝0.042-0.2。
In the fourth step, the process is as follows:
step 401, extracting a drill rod unloading video needing to be detected in a time period from storage equipment;
step 402, performing frame reduction processing on a video to be detected, and reducing the original video rate to 2 frames/s to read a video image;
step 403, normalizing the single-frame image into a 224 × 224 image, and inputting the normalized image into a trained adaptive learning rate ResNet-50 model;
step 404, displaying the classification confidence percentage of the output single-frame image on the relative image, and storing the confidence percentage into a CSV file based on a time sequence by taking a day as a unit;
and displaying the classification confidence percentage of the output single-frame image on the image, displaying the detection result of the single-frame image at the upper left corner of the image, displaying the detection result of the single-frame image as a Drill pipe + confidence percentage in a first category, displaying the detection result of the single-frame image as a No Drill pipe + confidence percentage in a second category, and storing the detection confidence result into a CSV file.
In step 5, the formula is applied during filtering
Wherein m is equal to N + and is less than or equal to 5 (m + 1) total frame number of the input video, f (x) is the CSV file data, when ∈ N + is greater than or equal to>
If yes, the 5 frame image is in the rod unloading process, the 5 frame is recorded as 1, and when->
Then the 5 frames are non-rodding processes and are recorded as 0.
The invention has the beneficial effects that:
(1) Acquiring a drill rod unloading video through original underground monitoring drilling equipment of a coal mine enterprise by an image processing technology, storing and frame-dismantling the drill rod unloading video, carrying out normalization processing on the images and storing all single-frame images; establishing a binary data set, dividing the binary data set into corresponding categories of the database according to the content of a single-frame image, fully and effectively utilizing original video resources of industrial and mining enterprises, reducing resource cost, establishing data which is most suitable for local use aiming at different regional backgrounds, and having irreplaceability and non-transportability;
(2) By training an improved self-adaptive learning rate ResNet-50 network, introducing a Logistic empirical formula on the basis of learning rate attenuation
Satisfies the equation-2 a ≈ nb, yields ≈ er>
The formula is shown in the specification, wherein n is the total number of learning rounds, the accuracy of the original model is improved, the accuracy of model training is improved to a greater extent, the identification precision is high, the model training is not easily influenced by shielding, and the using effect is good;
(3) The method has high detection precision, all images pass through a self-adaptive learning rate ResNet-50 model, the confidence coefficients of all second-class non-rod-unloading results in the CSV file are cleared, the integral method is used for filtering the confidence coefficient of video output, and a formula is used
Wherein m belongs to N + and is less than or equal to 5 (m + 1) total frame number of the input video, f (x) is CSV file data, when
The 5 frame image is taken as a knock-out process and the 5 frame is recorded as a 1 when->
The 5 frames are in a non-rod-unloading process and recorded as 0, the confidence coefficient is converted into signals of 0 and 1, errors caused by model detection can be effectively reduced, the final detection precision is improved, and meanwhile the problem of detection precision reduction caused by shielding is avoided.
Drawings
FIG. 1 is a flow chart of a visual statistical method for downhole drill pipe counting of the present invention;
FIG. 2 is a schematic diagram of a network used in a visual statistical method for downhole drill pipe counting according to the present invention;
FIG. 3 is a flow chart of a method of training an improved network for use in a visual statistics method of downhole drill pipe counting according to the present invention;
FIG. 4 is a flow chart of a method for counting the number of drill rods in the visual statistical method for downhole drill rod counting according to the invention.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
The invention relates to a visual statistical method for underground drill rod counting, which is implemented according to the following steps as shown in figure 1:
step one, data set collection and division: acquiring a drill rod unloading video through original underground monitoring drilling equipment of a coal mine enterprise, storing and framing the drill rod unloading video, carrying out normalization processing on images and storing all single-frame images; establishing a two-classification data set, and classifying the two-classification data set into corresponding classes of the database according to the content of the single-frame image;
the collected rod unloading video image comprises the complete rod unloading action of an operator and a drilling machine head; when the frame is disassembled, the frame is disassembled at the speed of one image per second for the drill rod disassembling video; the rod unloading video mainly selects time as 00: 00-2: 00, obtaining more than 10 ten thousand images of the initial training sample, renaming the images by the number of 1-100000, and avoiding the confusion of the classification caused by excessive images;
establishing a binary data set, a first type data set S 1 A second type data set S for the unloading drill rod category 2 For the non-unloading drill rod category, manually classifying the single-frame images, and setting the proportion of training data to test data to be 8 in a classification data set: 2, classifying the image data volume in the data set to be not less than 2000;
during normalization, the original video is subjected to frame splitting and batch normalization to be 224 × 224 images; classifying the normalized image into a discharging drill rod category S 1 And non-removable drill pipe category S 2 Generally, the image at the moment of non-drill rod unloading is far larger than the image of the drill rod being unloaded, the image at the moment of non-drill rod unloading needs to be randomly deleted, and the number ratio of the images in the two categories before deletion processing is S 1 :S 2 1 1 :S 2 =1, reducing data set imbalance, and ensuring that the total amount of data is not less than 2000 sheets;
the number of pixels of the data set acquisition camera is not less than 200 ten thousand;
step two, preprocessing a data set: for the unbalanced data quantity and insufficient data quantity of the two types of data in the data set, the data set quantity is increased by a preprocessing method combining rotation, turnover and brightness enhancement;
rotating, turning over and enhancing brightness, wherein the contrast and brightness of an image are enhanced in a Gamma conversion mode, and the method is mainly divided into six preprocessing means of only rotating, only turning over, only enhancing brightness, firstly enhancing the brightness and then rotating, firstly enhancing the brightness and then turning over, and firstly rotating and then reversing to increase a data set;
to unloading drill rod data S 1 Increasing a data set through preprocessing, wherein 10% of drill rod unloading data is subjected to rotation operation, 10% of drill rod unloading data is subjected to turnover operation, 20% of drill rod unloading data is subjected to brightness enhancement before rotation, 20% of drill rod unloading data is subjected to brightness enhancement before turnover, and 20% of drill rod unloading data is subjected to rotation before turnover;
the brightness enhancement adopts Gamma change, the Gamma transformation power parameter is set to be 0.5, the increased parameter is 10, the phenomenon of the unbalance of the two types of data of the data set is avoided, and the data quantity ratio of the two types of data sets is approximately equal to 1 after the data set is preprocessed: 1;
step three, training the improved adaptive learning rate ResNet-50 network, and the process is as follows:
step 301, extracting a data set and inputting the data set into a ResNet-50 network, wherein as shown in FIG. 2, the ResNet-50 network structure is composed of convolution 1, convolution 2, convolution 3, convolution 4, convolution 5, average pooling, full connection and a classifier;
a ResNet-50 network structure, wherein the step length of convolution 1 is 2, and the size of convolution kernel is 7 × 7; convolution 2 consists of 3 residual blocks; convolution 3 consists of 4 residual blocks; convolution 4 consists of 6 residual blocks; convolution 5 consists of 3 residual blocks; the activation function of the average pooling is softmax;
the residual block is composed of three layers, and is output to the next layer of residual block after 1 × 1 convolution, 3 × 3 convolution and 1 × 1 convolution; the activation function is Relu;
step 302, as shown in fig. 3, introducing Logistic empirical formula based on learning rate attenuation, at
Satisfies the equation-2 a ≈ nb, yields ≈ er>
A formula is shown, wherein n is the total number of learning rounds, and the accuracy of the original model is improved;
the Logistic empirical formula meets the condition that equation-2 a is approximately equal to nb, and the value of a is general
n is more than or equal to 50, the accuracy of the model can be effectively improved, n =100-120, b =0.042-0.2;
introducing a Logistic empirical formula of learning rate attenuation on the basis of an original Adam optimization function, wherein the change of a controls the horizontal translation of an attenuation curve, when a is increased, the attenuation curve moves towards the opposite direction of an x axis, otherwise, the attenuation curve moves towards the positive direction of the x axis, the change of b is related to the steepness or smoothness of the attenuation curve, when b is increased, the attenuation curve is steep, otherwise, the attenuation curve is smooth;
the model trained by the training set is brought into a test set, the classification precision of the model under the test set is detected, and the model is stored; a. b changes affect the test accuracy of the classification model, the center of the Logistic attenuation curve is fixed, namely the attenuation curve passes through (n/2, 1/2) points, and when b is increased from infinitesimal to b
During the interval, the testing precision of the classification model is gradually increased, b is continuously increased, and the testing precision of the classification model is gradually reduced; when a is taken from 0 as the starting value, is selected>
When a is increased, the test precision of the classification model begins to fluctuate and rise, when a meets-2 a ≈ nb, a is continuously increased, the test precision of the classification model gradually fluctuates and falls, and the result is that when a is increased, the test precision of the classification model begins to fluctuate and fall
In time, the accuracy of the classification model can be kept in a relatively high interval; eight minesWhen a = -4 and b = -0.1 are selected, the image classification detection precision can reach 0.877, and compared with the classification detection precision of 0.818 obtained without the Logistic attenuation curve, the data detection precision of the model is improved.
Inputting a rod unloading video, detecting the type of each frame of image in the video, outputting a confidence percentage, and storing the confidence percentage into a CSV file from an image time sequence; the process is as follows:
step 401, extracting a drill pipe unloading video needing to be detected in a time period from storage equipment;
step 402, performing frame reduction processing on a video to be detected, and reducing the original video rate to 2 frames/s to read a video image;
step 403, normalizing the single-frame image into a 224 × 224 image, and inputting the normalized image into a trained adaptive learning rate ResNet-50 model;
step 404, displaying the classification confidence percentage of the output single-frame image on the relative image, and storing the confidence percentage into a CSV file based on a time sequence by taking a day as a unit;
displaying the classification confidence percentage of the output single-frame image on the image, displaying the detection result of the single-frame image at the upper left corner of the image, displaying the detection result of the single-frame image as the percentage of Drill pipe + confidence, displaying the detection result of the single-frame image as the percentage of No Drill pipe + confidence, and storing the detection confidence result into a CSV file.
Step five, as shown in fig. 4, filtering by an integration method: after all images pass through a self-adaptive learning rate ResNet-50 model, clearing all confidence coefficients of the second class non-rod-unloading results in the CSV file, filtering the confidence coefficient of video output by using an integral method, and using a formula
Wherein m is equal to N + and is less than or equal to 5 (m + 1) total frame number of the input video, f (x) is the CSV file data, when ∈ N + is greater than or equal to>
If yes, the 5 frame image is in the rod unloading process, the 5 frame is recorded as 1, and when->
The 5 frames are in a non-rod-unloading process, recorded as 0, and the confidence coefficient is converted into signals of 0 and 1;
step six, counting the number of the drill rods: and counting the falling edges of the filtered 0 and 1 signals, and calculating the rod unloading number in the video through the falling edges.
And (4) counting the falling edges of the signals of 0 and 1, starting from the image data of the first frame of the CSV when the filtering data are arranged in a time sequence, counting the number +1 if the filtering data are changed from 1 to 0, and finally counting the total number of the drill rods.
Counting the number of the drill rods, setting a parameter C, if the number of frames between two falling edges is less than the parameter C, subtracting one from the number of the drill rods, otherwise, counting the number of the drill rods according to the original counting method if the number of frames between the two falling edges is greater than the parameter C.
By taking a flat coal eight-mine underground drill-unloading image as an example, an Adam optimization function is selected, when Logistic selects a = -4 and b = -0.1, the image classification detection precision can reach 0.877, and compared with the original classification precision of 0.818, the precision can be improved by 5.9%; aiming at the classification detection result, an integral method is selected for filtering to filter the original training result, so that the false detection condition caused by the error of the classification model is reduced; the method has the average detection precision higher than 97%, can count the rod unloading number at the rod unloading time in a non-contact mode, assists in evaluating whether the drilling length reaches the standard, reduces potential hazards caused by the fact that the drilling length does not reach the standard, and reduces the work labor intensity of counting workers.
Claims (6)
1. A visual statistical method for downhole drill pipe counting is characterized by being implemented according to the following steps:
collecting a drill rod unloading video, storing the drill rod unloading video, dismantling frames, carrying out normalization processing on images and storing all single-frame images; establishing a two-classification data set, and classifying the two-classification data set into corresponding classes of the database according to the content of the single-frame image;
step two, for unbalanced data quantity and insufficient data quantity of two types in the data set, increasing the data set quantity by a preprocessing method combining rotation, turnover and brightness enhancement;
step three, training an improved adaptive learning rate ResNet-50 network;
inputting a rod unloading video, detecting the image type of each frame in the video and outputting a confidence coefficient percentage; storing the confidence percentage from the image time sequence to a CSV file;
fifthly, after all the images pass through a self-adaptive learning rate ResNet-50 model, clearing the confidence coefficients of all the second non-rod-unloading results in the CSV file, filtering the video output confidence coefficient by using an integral method, and converting the confidence coefficients into signals of 0 and 1;
step six, counting the falling edges of the filtered 0 and 1 signals, and calculating the rod unloading number in the video through the falling edges;
and (5) counting the falling edges of the signals of 0 and 1, counting the number +1 if the filtering data is changed from 1 to 0, and finally counting the total number of the drill rods.
2. The visual statistical method for downhole drill pipe counting according to claim 1, wherein in the step 1, a binary data set is established, wherein the first type data set S 1 A second type data set S for the unloading drill rod category 2 For the non-unloading drill rod category, classifying the single-frame images manually, and setting the proportion of training data to test data in a classification data set as 8:2, classifying the image data volume in the data set to be not less than 2000; during normalization, the original video split frames are normalized in batches into images with the size of 224 × 224; classifying the normalized image into a discharging drill rod category S 1 And non-discharging drill rod class S 2 The image at the moment of non-drill rod unloading is far greater than the image of the drill rod being unloaded, the image at the moment of non-drill rod unloading needs to be randomly deleted, and the number ratio of the images in the two categories before deletion processing is S 1 :S 2 Down to S =1 1 :S 2 And 2, reducing data set unbalance and ensuring that the total data amount is not less than 2000 sheets.
3. A visual statistical method for downhole drill pipe counting according to claim 2, wherein in the step 2, drill pipe unloading data S is processed 1 By passingPreprocessing and increasing a data set, wherein 10% of the drill rod unloading data is subjected to rotation operation, 10% of the drill rod unloading data is subjected to turnover operation, 20% of the drill rod unloading data is subjected to brightness enhancement before rotation, 20% of the drill rod unloading data is subjected to brightness enhancement before turnover, and 20% of the drill rod unloading data is subjected to rotation before turnover; the brightness enhancement adopts Gamma change, the Gamma conversion power parameter is set to be 0.5, and the added parameter is 10.
4. The visual statistical method for downhole drill pipe counting according to claim 1, wherein in the third step, specifically:
step 301, extracting a data set and inputting the data set into a ResNet-50 network, wherein the ResNet-50 network structure is composed of a convolution 1, a convolution 2, a convolution 3, a convolution 4, a convolution 5, an average pooling, full connection and a classifier;
a ResNet-50 network structure, wherein the step size of convolution 1 is 2, and the size of a convolution kernel is 7 x 7; convolution 2 consists of 3 residual blocks; convolution 3 consists of 4 residual blocks; convolution 4 consists of 6 residual blocks; convolution 5 consists of 3 residual blocks; the activation function of the average pooling is softmax;
the residual block is composed of three layers, and is output to the next layer of residual block after 1 × 1 convolution, 3 × 3 convolution and 1 × 1 convolution; the activation function is Relu;
step 302, introducing a Logistic empirical formula on the basis of the attenuation of the learning rate
Satisfies the equation-2 a ≈ nb, yields ≈ er>
A formula, wherein n is the total number of learning rounds; the Logistic empirical formula satisfies the equation-2 a ≈ nb, and the value of a is ≈>
n＝100-120，b＝0.042-0.2。
5. The visual statistical method for downhole drill pipe counting according to claim 1, wherein in the fourth step, the process is as follows:
step 401, extracting a drill pipe unloading video needing to be detected in a time period from storage equipment;
step 402, performing frame reduction processing on a video to be detected, and reducing the original video rate to 2 frames/s to read a video image;
step 403, normalizing the single-frame image into a 224 × 224 image, and inputting the normalized image into a trained adaptive learning rate ResNet-50 model;
step 404, displaying the classification confidence percentage of the output single-frame image on the relative image, and storing the confidence percentage into a CSV file based on a time sequence by taking a day as a unit;
and displaying the classification confidence percentage of the output single-frame image on the image, displaying the detection result of the single-frame image at the upper left corner of the image, displaying the detection result of the single-frame image as a Drill pipe + confidence percentage in a first category, displaying the detection result of the single-frame image as a No Drill pipe + confidence percentage in a second category, and storing the detection confidence result into a CSV file.
6. A visual statistical method for downhole drill pipe counting according to claim 1, wherein in the step 5, a formula is applied during filtering
Wherein m is equal to N + and 5 (m + 1) is equal to or less than the total number of frames in the input video, f (x) is CSV file data, when ≥ N>
If yes, the 5 frame image is in the rod unloading process, the 5 frame is recorded as 1, and when->
Then the 5 frames are non-rodding processes and are recorded as 0./>
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010448311.XA CN111598042B (en) | 2020-05-25 | 2020-05-25 | Visual statistical method for underground drill rod counting |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010448311.XA CN111598042B (en) | 2020-05-25 | 2020-05-25 | Visual statistical method for underground drill rod counting |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111598042A CN111598042A (en) | 2020-08-28 |
| CN111598042B true CN111598042B (en) | 2023-04-07 |
Family
ID=72184031
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010448311.XA Active CN111598042B (en) | 2020-05-25 | 2020-05-25 | Visual statistical method for underground drill rod counting |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111598042B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112883830B (en) * | 2021-01-29 | 2022-03-15 | 南京北路智控科技股份有限公司 | Real-time automatic counting method for drill rods |
| CN113569658B (en) * | 2021-07-05 | 2024-02-20 | 天地（常州）自动化股份有限公司 | Intelligent management method for drilling mine based on video identification |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101710465A (en) * | 2009-12-16 | 2010-05-19 | 西南石油大学 | Method for simulating drilling tool lifting for drilling simulator |
| WO2019237567A1 (en) * | 2018-06-14 | 2019-12-19 | 江南大学 | Convolutional neural network based tumble detection method |
| CN110866512A (en) * | 2019-11-21 | 2020-03-06 | 南京大学 | Monitoring camera shielding detection method based on video classification |
-
2020
- 2020-05-25 CN CN202010448311.XA patent/CN111598042B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101710465A (en) * | 2009-12-16 | 2010-05-19 | 西南石油大学 | Method for simulating drilling tool lifting for drilling simulator |
| WO2019237567A1 (en) * | 2018-06-14 | 2019-12-19 | 江南大学 | Convolutional neural network based tumble detection method |
| CN110866512A (en) * | 2019-11-21 | 2020-03-06 | 南京大学 | Monitoring camera shielding detection method based on video classification |
Non-Patent Citations (1)
| Title |
|---|
| 近岸海浪视频浪高自动检测;宋巍等;《中国图象图形学报》(第03期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111598042A (en) | 2020-08-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110852395B (en) | Ore granularity detection method and system based on autonomous learning and deep learning | |
| CN111598042B (en) | Visual statistical method for underground drill rod counting | |
| CN109272123B (en) | Sucker-rod pump working condition early warning method based on convolution-circulation neural network | |
| CN113283395B (en) | Video detection method for blocking foreign matters at transfer position of coal conveying belt | |
| CN110163278A (en) | A kind of flame holding monitoring method based on image recognition | |
| CN111127448B (en) | Method for detecting air spring fault based on isolated forest | |
| CN112508944A (en) | Leakage detection method applied to underground water supply pipeline of coal mine | |
| CN113989257A (en) | Electric power comprehensive pipe gallery settlement crack identification method based on artificial intelligence technology | |
| CN111402236A (en) | Hot-rolled strip steel surface defect grading method based on image gray value | |
| CN106872473A (en) | A kind of potato defects detection identifying system design based on machine vision | |
| CN117576632B (en) | Multi-mode AI large model-based power grid monitoring fire early warning system and method | |
| CN106447656A (en) | Rendering flawed image detection method based on image recognition | |
| CN113469938B (en) | Pipe gallery video analysis method and system based on embedded front-end processing server | |
| CN114581666A (en) | Intelligent rock sample identification method and system based on image deep learning | |
| CN110555384A (en) | Beef marbling automatic grading system and method based on image data | |
| CN114387564A (en) | Head-knocking engine-off pumping-stopping detection method based on YOLOv5 | |
| CN113610831B (en) | Wood defect detection method based on computer image technology and transfer learning | |
| CN112668634B (en) | Underground pipe network automatic identification and detection cloud system based on deep learning | |
| CN112990350B (en) | Target detection network training method and target detection network-based coal and gangue identification method | |
| CN113298148B (en) | Ecological environment evaluation-oriented unbalanced data resampling method | |
| CN112730437B (en) | Spinneret plate surface defect detection method and device based on depth separable convolutional neural network, storage medium and equipment | |
| CN111784059B (en) | Method for predicting dominant development azimuth of coal seam macroscopic crack | |
| CN114004151A (en) | Oil pumping well fault prediction method based on LSTM-VGGNet | |
| CN113935417A (en) | Method for counting underground drill rods based on whole-process analysis | |
| CN112907734A (en) | TEDS fault detection method based on virtual CRH380A model and deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Du Jingyi Inventor after: Zhang Houbin Inventor after: Liu Binchao Inventor after: Hao Le Inventor after: Liu Zhanqing Inventor after: Du Jin Inventor after: Cui Lijun Inventor after: Chen Rui Inventor after: Shi Zhimang Inventor after: Chen Yuhang Inventor after: Dong Gang Inventor before: Du Jingyi Inventor before: Hao Le Inventor before: Chen Rui Inventor before: Shi Zhimang Inventor before: Chen Yuhang Inventor before: Dong Gang Inventor before: Zhang Houbin Inventor before: Liu Binchao |