CN115565020A - Tooth surface damage identification method and device based on improved neural network - Google Patents
Tooth surface damage identification method and device based on improved neural network Download PDFInfo
- Publication number
- CN115565020A CN115565020A CN202211414681.7A CN202211414681A CN115565020A CN 115565020 A CN115565020 A CN 115565020A CN 202211414681 A CN202211414681 A CN 202211414681A CN 115565020 A CN115565020 A CN 115565020A
- Authority
- CN
- China
- Prior art keywords
- dimensional feature
- feature map
- carrying
- tooth surface
- dimensional
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 28
- 230000006870 function Effects 0.000 claims abstract description 67
- 238000012549 training Methods 0.000 claims abstract description 25
- 238000011176 pooling Methods 0.000 claims abstract description 14
- 238000013145 classification model Methods 0.000 claims abstract description 12
- 238000013526 transfer learning Methods 0.000 claims abstract description 5
- 238000010606 normalization Methods 0.000 claims description 72
- 238000010586 diagram Methods 0.000 claims description 59
- 230000004913 activation Effects 0.000 claims description 46
- 238000013507 mapping Methods 0.000 claims description 46
- 239000011159 matrix material Substances 0.000 claims description 36
- 238000013527 convolutional neural network Methods 0.000 claims description 33
- 238000012360 testing method Methods 0.000 claims description 12
- 239000013598 vector Substances 0.000 claims description 12
- 238000013138 pruning Methods 0.000 claims description 8
- 238000004458 analytical method Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 5
- 238000005299 abrasion Methods 0.000 claims description 4
- 238000004590 computer program Methods 0.000 claims description 4
- 238000003860 storage Methods 0.000 claims description 4
- 235000002566 Capsicum Nutrition 0.000 claims description 3
- 239000006002 Pepper Substances 0.000 claims description 3
- 235000016761 Piper aduncum Nutrition 0.000 claims description 3
- 235000017804 Piper guineense Nutrition 0.000 claims description 3
- 235000008184 Piper nigrum Nutrition 0.000 claims description 3
- 238000005282 brightening Methods 0.000 claims description 3
- 238000005520 cutting process Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 claims description 3
- 150000003839 salts Chemical class 0.000 claims description 3
- 238000004804 winding Methods 0.000 claims description 3
- 244000203593 Piper nigrum Species 0.000 claims 1
- 238000003062 neural network model Methods 0.000 abstract description 2
- 238000005457 optimization Methods 0.000 description 9
- 230000002093 peripheral effect Effects 0.000 description 3
- 241000722363 Piper Species 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000005452 bending Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000009347 mechanical transmission Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
A tooth surface damage identification method and a tooth surface damage identification device based on an improved neural network belong to the field of tooth surface damage identification of gears. According to the method, a neural network model is improved, the global average pooling is used for adapting to the tooth surface size and accelerating training, and the model is converged more quickly by applying transfer learning; and finally, loading the data set into an improved ResNet-34 to extract and train image features, classifying by using a classification function to finally obtain a tooth surface damage classification model, and inputting the gear surface image to be detected into the model, so that the precision of damage identification can be ensured, the model is effectively compressed, the training cost is reduced, and the identification efficiency is greatly improved.
Description
Technical Field
The invention relates to the field of tooth surface damage identification of gears, in particular to a tooth surface damage identification method and device based on an improved neural network.
Background
Gears are key components of mechanical transmission systems, the quality of which directly affects the operating conditions and the service life of the transmission system. However, due to manufacturing and installation errors, motion impact, sliding friction, cyclic alternating bending stress, material properties and heat treatment processes, different failure modes of the gear may occur, and the failure modes are commonly represented by tooth surface pitting, tooth surface abrasion, tooth breakage and the like. The damage of the tooth surface can cause vibration and noise, reduce the service life and influence the motion precision and the working stability of the gear.
At present, the identification of the tooth surface damage type mainly depends on the visual detection of workshop workers and maintenance personnel, and has low efficiency, poor real-time performance and higher requirement on the workers. With the rapid development of computer technology and artificial intelligence, machine identification gradually replaces manual operation, but at present, no unified standard or method exists at home and abroad for identifying the tooth surface damage type of the gear.
Disclosure of Invention
The invention aims to provide a tooth surface damage identification method based on an improved neural network, which adopts a machine vision system to replace manual operation, and improves the accuracy and efficiency of gear tooth surface damage identification.
The technical scheme adopted by the invention for realizing the technical purpose is as follows: a tooth surface damage identification method based on an improved neural network is characterized in that the improved neural network is utilized to establish a tooth surface damage classification model, and then a gear surface image to be detected is input into the tooth surface damage classification model to obtain an analysis result of the gear surface damage; the steps of establishing the tooth surface damage classification model by utilizing the improved neural network are as follows:
firstly), acquiring a tooth surface image of a gear by using a machine vision system, classifying the acquired tooth surface image according to the characteristics of tooth surface damage, and marking a real label of each classification;
secondly), cutting the tooth surface images of each type to the same size, and performing image enhancement to form a classified tooth surface image data set;
thirdly), constructing a convolutional neural network by taking ResNet-34 as a primary network model, training the constructed convolutional neural network by utilizing a tooth surface image data set, and storing formed weights into the constructed convolutional neural network;
fourthly) carrying out classification processing on the output result of the convolutional neural network by using a classification function softmax to obtain probability vectors of each classification;
and fifthly), selecting the maximum value in the probability vector as a prediction type, and outputting the prediction type and the prediction probability as an analysis result of the gear surface damage.
As an optimization scheme of the tooth surface damage identification method based on the improved neural network, the step one) of classifying the acquired tooth surface images means that the acquired tooth surface images are classified according to four types of tooth surface pitting, tooth surface abrasion, gear tooth breakage and normal tooth surface, and the four types are used as real labels for corresponding classification.
As another optimization scheme of the tooth surface damage identification method based on the improved neural network, the image enhancement in the step two) is sequentially performed by horizontal turning, vertical turning, 180-degree rotation, zooming, shifting, blurring, brightening, darkening and salt and pepper noise adding.
As another optimization scheme of the tooth surface damage identification method based on the improved neural network, in the second step), a classified tooth surface image data set is formed, and the proportion of the training set to the test set is 9:1.
as another optimization scheme of the tooth surface damage identification method based on the improved neural network, the convolutional neural network in the step three) sequentially comprises an input layer, a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, a fifth convolutional layer, a global average pooling layer and a full connection layer.
As another optimization scheme of the tooth surface damage identification method based on the improved neural network, the specific operation of constructing the convolutional neural network in the step three) is as follows:
1) The first winding layer
1.1 Passing the image with the dimension of (224, 64) through 64 convolution kernels of 7 by 7 and performing convolution operation with the step size of 2, and performing zero filling operation for 3 circles at the periphery of the obtained pixel matrix to obtain a three-dimensional characteristic map of (112, 64);
1.2 Carrying out batch normalization operation on the obtained three-dimensional characteristic graph, and then carrying out nonlinear mapping operation on the three-dimensional characteristic graph subjected to batch normalization through a ReLU activation function;
1.3 Maximum pooling operation of the signature output from step 1.2) by 3 × 3 convolution kernels and in 2 steps, followed by 1 cycle of zero padding operation at the periphery of the resulting pixel matrix to obtain a three-dimensional signature (56, 64);
2) The second convolution layer
2.1 Carrying out convolution on the three-dimensional feature map (56, 64) obtained in the previous step by 64 convolution kernels of 3 x 3 and with the step size of 1, and carrying out 1-turn zero padding operation on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map (56, 64);
2.2 Batch normalization operation is carried out on the three-dimensional feature maps (56, 64) obtained in the previous step, and then nonlinear mapping operation is carried out on the batch normalized three-dimensional feature maps through a ReLU activation function;
2.3 ) repeating the convolution and zero filling operation in the step 2.1) on the three-dimensional characteristic diagram obtained in the previous step, and carrying out batch normalization operation on the obtained three-dimensional characteristic diagram;
2.4 Adding the three-dimensional characteristic diagram input in the step 2.1) and the three-dimensional characteristic diagram output in the step 2.3), and then carrying out nonlinear mapping operation on the three-dimensional characteristic diagram obtained after addition through a ReLU activation function;
2.5 Repeating the step 2.1) to the step 2.4) twice, and outputting a three-dimensional feature map with dimensions (56, 64);
3) The third convolution layer
3.1 The three-dimensional characteristic map with the dimension of (56, 64) obtained in the previous step is convoluted by 128 convolution kernels of 3 x 3 and step size of 2, and zero padding operation is carried out for 1 circle at the periphery of the obtained pixel matrix to obtain a three-dimensional characteristic map of (28, 128);
3.2 Carrying out batch normalization operation on the three-dimensional feature maps (28, 128) obtained in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature maps after batch normalization through a ReLU activation function to obtain three-dimensional feature maps with the dimension of (28, 64);
3.3 Carrying out convolution on the three-dimensional feature map with the dimension of (28, 64) obtained in the previous step through 128 convolution kernels of 3 x 3 and the step size of 1, carrying out zero filling operation for 1 circle on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map with the dimension of (28, 128), and carrying out batch normalization operation on the three-dimensional feature map;
3.4 Carrying out convolution on the three-dimensional feature map input in the step 3.1) through 128 convolution kernels of 1 × 1 and with the step size of 2 to obtain a (28, 128) three-dimensional feature map, adding the three-dimensional feature map and the feature map obtained in the step 3.3), and then carrying out nonlinear mapping operation on the three-dimensional feature map obtained after addition through a ReLU activation function to obtain a three-dimensional feature map with the dimension of (28, 128);
3.5 Passing the three-dimensional feature map (28, 128) obtained in the previous step through 128 convolution kernels of 3 × 3 and performing convolution with a step size of 1, and performing zero padding operation for 1 circle on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map (28, 128);
3.6 Carrying out batch normalization operation on the three-dimensional characteristic diagram output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional characteristic diagram after batch normalization through a ReLU activation function;
3.7 Step 3.5) is repeated and the obtained three-dimensional characteristic diagram is subjected to batch normalization operation;
3.8 Adding the three-dimensional feature map output in the previous step with the three-dimensional feature map with the dimension (28, 128) input in the step 3.5), and then carrying out nonlinear mapping operation on the three-dimensional feature map obtained after addition through a ReLU activation function;
3.9 Repeating the step 3.5) to the step 3.8) twice, and outputting a three-dimensional feature map with the dimension of (28, 128);
4) The fourth convolution layer
4.1 Passing the three-dimensional feature map obtained in the previous step through 256 convolution kernels of 3 x 3 and performing convolution by 2 steps, and performing zero padding operation for 1 circle on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map (14, 256);
4.2 Carrying out batch normalization operation on the three-dimensional feature map (14, 256) obtained in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature map after batch normalization through a ReLU activation function to obtain a three-dimensional feature map with the dimension of (14, 256);
4.3 Passing the three-dimensional feature map obtained in the previous step through 256 convolution kernels of 3 x 3 and performing convolution by step size of 1, performing zero padding operation for 1 circle on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map (14, 256), and performing batch normalization operation on the three-dimensional feature map;
4.4 Carrying out convolution on the three-dimensional feature map input in the step 4.1) through 256 convolution kernels of 1 × 1 and in steps of 2, adding the three-dimensional feature map with the dimensionality (14, 256) and the three-dimensional feature map output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature map obtained after addition through a ReLU activation function;
4.5 Passing the three-dimensional feature map output in the previous step through 256 convolution kernels of 3 x 3 and performing convolution by step size of 1, and performing zero filling operation for 1 circle on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map (14, 256);
4.6 Carrying out batch normalization operation on the three-dimensional characteristic diagram output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional characteristic diagram after batch normalization through a ReLU activation function;
4.7 Step 4.5) is repeated and batch normalization operation is carried out on the obtained three-dimensional characteristic diagram;
4.8 Adding the three-dimensional characteristic diagram input in the step 4.5) with the three-dimensional characteristic diagram output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional characteristic diagram obtained after addition through a ReLU activation function;
4.9 Repeating the step 4.5) to the step 4.8) four times, and outputting a three-dimensional feature map with the dimension of (14, 256);
5) The fifth convolution layer
5.1 Carrying out convolution on the three-dimensional feature map obtained in the previous step by 512 convolution kernels of 3 x 3 and 2 steps, and carrying out 1-turn zero filling operation on the periphery of the obtained pixel matrix to obtain a (7, 512) three-dimensional feature map;
5.2 Carrying out batch normalization operation on the three-dimensional characteristic diagram obtained in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional characteristic diagram after batch normalization through a ReLU activation function to obtain a three-dimensional characteristic diagram with the dimensionality of (7, 512);
5.3 Carrying out convolution on the three-dimensional feature map obtained in the previous step by 512 convolution kernels of 3 x 3 and step length of 1, carrying out 1-turn zero padding operation on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map of (7, 512), and carrying out batch normalization operation on the three-dimensional feature map;
5.4 Carrying out convolution on the three-dimensional feature map input in the step 5.1) through 512 convolution kernels of 1 × 1 and with the step size of 2 to obtain a three-dimensional feature map with the dimensionality of (7, 512) and the three-dimensional feature map output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature map obtained after the addition through a ReLU activation function;
5.5 The three-dimensional feature map output in the previous step is convoluted by 512 convolution kernels of 3 x 3 and the step size of 1, and the zero filling operation is carried out for 1 circle at the periphery of the obtained pixel matrix, so as to obtain the three-dimensional feature map of (7, 512);
5.6 Carrying out batch normalization operation on the three-dimensional feature maps output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature maps subjected to batch normalization through a ReLU activation function;
5.7 Step 5.5) is repeated and batch normalization operation is carried out on the obtained three-dimensional characteristic diagram;
5.8 Adding the three-dimensional characteristic diagram input in the step 5.5) with the three-dimensional characteristic diagram output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional characteristic diagram obtained after addition through a ReLU activation function;
5.9 Repeating the step 5.5) to the step 5.8) once, and outputting a three-dimensional feature map with the dimension of (7, 512);
6) Global average pooling layer
Carrying out global average pooling operation on the three-dimensional feature map with the dimension (7, 512) output in the previous step, and outputting the three-dimensional feature map with the dimension (1, 512);
7) Full connection layer
And converting the three-dimensional feature map with the dimension of (1, 512) output in the previous step into a 4-dimensional feature vector.
As another optimization scheme of the tooth surface damage identification method based on the improved neural network, after batch normalization operations in step 1.2), step 2.2), step 2.3), step 3.2), step 3.3), step 3.6), step 3.7), step 4.2), step 4.3), step 4.6), step 4.7), step 5.2), step 5.3), step 5.6) and step 5.7), pruning operation with a pruning rate of 20% is performed on each batch normalization layer.
As another optimization scheme of the tooth surface damage identification method based on the improved neural network, the training of the constructed convolutional neural network and the saving of the formed weights in the constructed convolutional neural network comprise the following specific operations:
a) Inputting a training set in the tooth surface image data set into a built convolutional neural network for training, performing transfer learning by using official weight, and finally forming a plurality of output numerical values corresponding to the classification quantity;
b) Use of
A function, converting a plurality of output values into relative probabilities to obtain probability vectors corresponding to each classification, and selecting the classification with the highest probability as a predicted value of the output;
in the formula, i is a category index, C is a category number, and is the output of a preceding stage output unit of the classifier of the full connection layer, and e is a natural base number;
c) Inputting the obtained predicted value and the real label into a cross entropy function cross EntropyLoss to calculate an error value, calculating the gradient value of each parameter through an error back propagation function loss.
d) Inputting the test set into a convolutional neural network to verify the training effect, calculating an error value according to a cross entropy function, outputting the type with the maximum probability as a predicted value, judging that the prediction is correct if the predicted value is the same as the real label, and determining the ratio of the number of correct predictions to the number of test sets as the accuracy of the whole test set;
e) Training for 400 rounds, wherein the lowest error value on the set to be tested is 0.095, the accuracy rate reaches 97.7%, and the trained weight is stored in the convolutional neural network.
An electronic device comprising a processor and a memory, the memory storing computer readable instructions that, when executed by the processor, are capable of executing the above-described tooth flank damage identification method.
A readable storage medium, on which a computer program is stored which, when being executed by a processor, performs the steps of the above-mentioned tooth flank damage identification method.
Compared with the prior art, the invention has the following beneficial effects:
according to the method, a neural network model is improved, parameters and calculated amount are reduced by a channel pruning method within a range allowed by precision errors, global average pooling is used for adapting to tooth surface size and accelerating training, and the model is converged more quickly by applying transfer learning; and finally, loading the data set into an improved ResNet-34 to extract and train image features, classifying by using a classification function to finally obtain a tooth surface damage classification model, and inputting the gear surface image to be detected into the model, so that the precision of damage recognition can be ensured, the model is effectively compressed, the training cost is reduced, and the recognition efficiency is greatly improved.
Drawings
FIG. 1 is a schematic illustration of a tooth surface damage classification model processing a tooth surface image in accordance with the present invention;
Detailed Description
The technical solution of the present invention is further described in detail with reference to specific embodiments, and for example, for denoising, enhancing and edge detecting of an image, classification and identification by a classifier, batch normalization operation, nonlinear mapping by a ReLU activation function, pruning operation, and other operations used such as Softmax function, cross entropy function cross entropy, error back propagation function low, gradient descent function optimization step () are understood as the prior art known or should be known to those skilled in the art.
Example 1
A tooth surface damage identification method based on an improved neural network is disclosed, as shown in figure 1, the method utilizes the improved neural network to establish a tooth surface damage classification model, and then obtains an analysis result of the surface damage of a gear by inputting a gear surface image to be detected into the tooth surface damage classification model; the steps of establishing the tooth surface damage classification model by utilizing the improved neural network are as follows:
firstly), acquiring a tooth surface image of a gear by using a machine vision system, classifying the acquired tooth surface image according to the characteristics of tooth surface damage, and labeling a real label of each classification;
in practice, the damage identification and recognition can be carried out manually, and classification and real labels are marked;
classifying the acquired tooth surface images in the step means that the acquired tooth surface images are classified according to four types of tooth surface pitting, tooth surface abrasion, gear tooth breakage and normal tooth surface, and the four types are used as real labels of corresponding classification;
secondly), cutting the tooth surface images of each type to the same size, and performing image enhancement to form a classified tooth surface image data set;
in the step, the image enhancement method comprises the steps of horizontal turning, vertical turning, 180-degree rotation, zooming, shifting, blurring, brightening, darkening and adding salt and pepper noise in sequence;
in the tooth surface image data set for which classification is formed in this step, the ratio of the training set to the test set is 9:1;
thirdly), constructing a convolutional neural network by taking ResNet-34 as a primary network model, training the constructed convolutional neural network by utilizing a tooth surface image data set, and storing formed weights into the constructed convolutional neural network;
the convolutional neural network in the step sequentially comprises an input layer, a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, a fifth convolutional layer, a global average pooling layer and a full-connection layer;
the specific operation of constructing the convolutional neural network in this step is, as shown in fig. 1, as follows:
1) The first winding layer
1.1 Passing the image with the dimension of (224, 64) through 64 convolution kernels of 7 by 7 and performing convolution operation with the step size of 2, and performing zero filling operation for 3 circles at the periphery of the obtained pixel matrix to obtain a three-dimensional characteristic map of (112, 64);
1.2 Carrying out batch normalization operation on the obtained three-dimensional characteristic diagram, and then carrying out nonlinear mapping operation on the batch normalized three-dimensional characteristic diagram through a ReLU activation function;
1.3 Maximum pooling operation of the signature output from step 1.2) by 3 × 3 convolution kernels and in 2 steps, followed by 1 cycle of zero padding operation at the periphery of the resulting pixel matrix to obtain a three-dimensional signature (56, 64);
2) A second convolution layer
2.1 The three-dimensional characteristic diagram (56, 64) obtained in the previous step is convoluted by 64 convolution kernels of 3 x 3 and with the step size of 1, and the zero padding operation is carried out for 1 circle on the periphery of the obtained pixel matrix to obtain the three-dimensional characteristic diagram (56, 64);
2.2 Carrying out batch normalization operation on the three-dimensional feature maps (56, 56 and 64) obtained in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature maps after batch normalization through a ReLU activation function;
2.3 ) repeating the convolution and zero filling operation in the step 2.1) on the three-dimensional characteristic diagram obtained in the previous step, and carrying out batch normalization operation on the obtained three-dimensional characteristic diagram;
2.4 Adding the three-dimensional characteristic diagram input in the step 2.1) and the three-dimensional characteristic diagram output in the step 2.3), and then carrying out nonlinear mapping operation on the three-dimensional characteristic diagram obtained after addition through a ReLU activation function;
2.5 Repeating the step 2.1) to the step 2.4) twice, and outputting a three-dimensional feature map with dimensions (56, 64);
3) The third convolution layer
3.1 Carrying out convolution on the three-dimensional feature map with the dimension (56, 64) obtained in the previous step through 128 convolution kernels with 3 x 3 and the step size of 2, and carrying out 1-turn zero padding operation on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map (28, 128);
3.2 Carrying out batch normalization operation on the three-dimensional feature maps (28, 128) obtained in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature maps after batch normalization through a ReLU activation function to obtain three-dimensional feature maps with the dimension of (28, 64);
3.3 Carrying out convolution on the three-dimensional feature map with the dimensionality of (28, 64) obtained in the previous step through 128 convolution kernels with 3 x 3 and the step size of 1, carrying out 1-turn zero filling operation on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map with (28, 128), and carrying out batch normalization operation on the three-dimensional feature map;
3.4 Carrying out convolution on the three-dimensional feature map input in the step 3.1) through 128 convolution kernels of 1 × 1 and with the step size of 2 to obtain a (28, 128) three-dimensional feature map, adding the three-dimensional feature map and the feature map obtained in the step 3.3), and then carrying out nonlinear mapping operation on the three-dimensional feature map obtained after addition through a ReLU activation function to obtain a three-dimensional feature map with the dimension of (28, 128);
3.5 Passing the three-dimensional feature map (28, 128) obtained in the previous step through 128 convolution kernels of 3 × 3 and performing convolution with a step size of 1, and performing zero padding operation for 1 circle on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map (28, 128);
3.6 Carrying out batch normalization operation on the three-dimensional characteristic diagram output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional characteristic diagram after batch normalization through a ReLU activation function;
3.7 Step 3.5) is repeated and batch normalization operation is carried out on the obtained three-dimensional characteristic diagram;
3.8 Adding the three-dimensional feature map output in the previous step with the three-dimensional feature map with the dimension (28, 128) input in the step 3.5), and then carrying out nonlinear mapping operation on the three-dimensional feature map obtained after addition through a ReLU activation function;
3.9 Repeating the step 3.5) to the step 3.8) twice, and outputting a three-dimensional feature map with the dimension of (28, 128);
4) The fourth convolution layer
4.1 Passing the three-dimensional feature map obtained in the previous step through 256 convolution kernels of 3 × 3 and performing convolution with 2 steps, and performing 1-turn zero padding operation on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map of (14, 256);
4.2 Carrying out batch normalization operation on the three-dimensional feature map (14, 256) obtained in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature map after batch normalization through a ReLU activation function to obtain a three-dimensional feature map with the dimension of (14, 256);
4.3 Passing the three-dimensional feature map obtained in the previous step through 256 convolution kernels of 3 x 3 and performing convolution by step size of 1, performing zero padding operation for 1 circle on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map (14, 256), and performing batch normalization operation on the three-dimensional feature map;
4.4 Carrying out convolution on the three-dimensional feature map input in the step 4.1) through 256 convolution kernels of 1 × 1 and in steps of 2, adding the three-dimensional feature map with the dimensionality (14, 256) and the three-dimensional feature map output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature map obtained after addition through a ReLU activation function;
4.5 Passing the three-dimensional feature map output in the previous step through 256 convolution kernels of 3 x 3 and performing convolution by step size of 1, and performing zero filling operation for 1 circle on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map (14, 256);
4.6 Carrying out batch normalization operation on the three-dimensional characteristic diagram output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional characteristic diagram after batch normalization through a ReLU activation function;
4.7 Step 4.5) is repeated and batch normalization operation is carried out on the obtained three-dimensional characteristic diagram;
4.8 Adding the three-dimensional feature map input in the step 4.5) with the three-dimensional feature map output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature map obtained after addition through a ReLU activation function;
4.9 Step 4.5) -step 4.8) are repeated four times, and a three-dimensional feature map with the dimensionality (14, 256) is output;
5) The fifth convolution layer
5.1 Passing the three-dimensional feature map obtained in the previous step through 512 convolution kernels of 3 x 3 and performing convolution by step size of 2, and performing zero padding operation for 1 circle on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map of (7, 512);
5.2 Carrying out batch normalization operation on the three-dimensional characteristic diagram obtained in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional characteristic diagram after batch normalization through a ReLU activation function to obtain a three-dimensional characteristic diagram with the dimensionality of (7, 512);
5.3 The three-dimensional feature map obtained in the previous step is convoluted by 512 convolution kernels of 3 x 3 and the step length of 1, and the zero padding operation is carried out for 1 circle on the periphery of the obtained pixel matrix to obtain the three-dimensional feature map of (7, 512), and the three-dimensional feature map is subjected to batch normalization operation;
5.4 Carrying out convolution on the three-dimensional feature map input in the step 5.1) through 512 convolution kernels of 1 × 1 and 2 steps to obtain a three-dimensional feature map with the dimension of (7, 512), adding the three-dimensional feature map with the dimension of (7, 512) with the three-dimensional feature map output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature map obtained after addition through a ReLU activation function;
5.5 The three-dimensional feature map output in the previous step is convoluted by 512 convolution kernels of 3 x 3 and the step size of 1, and the zero filling operation is carried out for 1 circle at the periphery of the obtained pixel matrix, so as to obtain the three-dimensional feature map of (7, 512);
5.6 Carrying out batch normalization operation on the three-dimensional characteristic diagram output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional characteristic diagram after batch normalization through a ReLU activation function;
5.7 Step 5.5) is repeated and the obtained three-dimensional characteristic diagram is subjected to batch normalization operation;
5.8 Adding the three-dimensional feature map input in the step 5.5) with the three-dimensional feature map output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature map obtained after addition through a ReLU activation function;
5.9 Repeating the step 5.5) to the step 5.8) once, and outputting a three-dimensional feature map with the dimension of (7, 512);
6) Global average pooling layer
Performing global average pooling operation on the three-dimensional feature map with the dimension of (7, 512) output in the previous step, and outputting the three-dimensional feature map with the dimension of (1, 512);
7) Full connection layer fc
Converting the three-dimensional feature map with the dimensionality of (1, 512) output in the previous step into a 4-dimensional feature vector;
fourthly) classifying the output result of the convolutional neural network by using a classification function softmax to obtain probability vectors of each classification;
and fifthly), selecting the maximum value in the probability vector as a prediction type, and outputting the prediction type and the prediction probability as an analysis result of the gear surface damage.
In this embodiment, after the batch normalization operations in step 1.2), step 2.2), step 2.3), step 3.2), step 3.3), step 3.6), step 3.7), step 4.2), step 4.3), step 4.6), step 4.7), step 5.2), step 5.3), step 5.6) and step 5.7), the pruning operation with the pruning rate of 20% is performed on each batch normalization layer.
In this embodiment, the specific operations of training the constructed convolutional neural network and storing the formed weights in the constructed convolutional neural network are as follows:
a) Inputting a training set in the tooth surface image data set into a built convolutional neural network for training, and performing transfer learning by using official weight to finally form a plurality of output numerical values corresponding to the classification quantity;
b) Use of
A function, wherein i is a category index, C is the number of categories and is the output of a preceding-stage output unit of a classifier of a full connection layer, e is a natural base number, a plurality of output values are converted into relative probabilities to obtain probability vectors corresponding to the categories, and the category with the highest probability is selected as a predicted value of the output;
c) Inputting the obtained predicted value and the real label into a cross entropy function CrossEntropyLoss to calculate an error value, calculating a gradient value of each parameter through an error back propagation function loss.
d) Inputting the test set into a convolutional neural network to verify the training effect, calculating an error value according to a cross entropy function, outputting the type with the maximum probability as a predicted value, and if the predicted value is the same as the real label, determining that the prediction is correct, wherein the ratio of the number of the correct prediction to the number of the test set is the accuracy of the whole test set;
e) Training for 400 times, wherein the lowest error value on the set to be tested is 0.095, the accuracy rate reaches 97.7%, and the trained weight is stored in the convolutional neural network.
Example 2
An electronic device comprising a processor and a memory, the memory storing computer readable instructions that, when executed by the processor, are capable of performing the optimization method of embodiment 1.
In this embodiment, the electronic apparatus includes a modeling device, a processor, a memory, a storage controller, a peripheral interface, an input-output unit, an audio unit, a display unit, and the like.
Specifically, the memory controller, the processor, the peripheral interface, the input/output unit, the audio unit, and the display unit are electrically connected to each other directly or indirectly to implement data transmission or interaction. The modelling means includes at least one software functional module which may be stored in said memory in the form of software or firmware (firmware) or fixed in an Operating System (OS) of the modelling means. The processor is used to execute executable modules stored in the memory, including software functional modules or computer programs.
The memory may be, but is not limited to, a Random Access Memory (RAM), a Read Only Memory (ROM), a Programmable Read Only Memory (PROM), an erasable read only memory (EPROM), an electrically erasable read only memory (EEPROM), and the like. The memory is used for storing programs, and the processor executes corresponding programs after receiving execution instructions.
The processor may be an integrated circuit chip having signal processing capabilities. The processor may be a general-purpose processor including a Central Processing Unit (CPU), a Network Processor (NP), etc.; but may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components.
The peripheral interface couples various input/output devices to the processor and memory, which may be implemented in a single chip or, alternatively, may be implemented separately from a separate chip.
The input and output unit, the audio unit and the display unit are all the prior art, for example, the input and output unit is used for providing input data for a user to realize the interaction between the user and the server (or the local terminal), and can be a mouse, a keyboard and the like; such as an audio unit, which may include one or more microphones, one or more speakers, and audio circuitry; such as a display unit, provides an interactive interface (e.g., a user interface) between the electronic device and a user or for displaying image data for reference by the user.
Example 3
A readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of embodiment 1.
Claims (10)
1. A tooth surface damage identification method based on an improved neural network is characterized in that the improved neural network is utilized to establish a tooth surface damage classification model, and then a gear surface image to be detected is input into the tooth surface damage classification model to obtain an analysis result of the gear surface damage; the method is characterized in that the steps of establishing the tooth surface damage classification model by utilizing the improved neural network are as follows:
firstly), acquiring a tooth surface image of a gear by using a machine vision system, classifying the acquired tooth surface image according to the characteristics of tooth surface damage, and marking a real label of each classification;
secondly), cutting the tooth surface images of each type to the same size, and performing image enhancement to form a classified tooth surface image data set;
thirdly), constructing a convolutional neural network by taking ResNet-34 as a primary network model, training the constructed convolutional neural network by utilizing a tooth surface image data set, and storing formed weights into the constructed convolutional neural network;
fourthly) carrying out classification processing on the output result of the convolutional neural network by using a classification function softmax to obtain probability vectors of each classification;
and fifthly), selecting the maximum value in the probability vector as a prediction type, and outputting the prediction type and the prediction probability as an analysis result of the gear surface damage.
2. The method for identifying the tooth surface damage based on the improved neural network as claimed in claim 1, wherein: the step one) of classifying the acquired tooth surface images means that the acquired tooth surface images are classified according to four types of tooth surface pitting, tooth surface abrasion, gear tooth breakage and normal tooth surface, and the four types are used as correspondingly classified real labels.
3. The method for identifying the tooth surface damage based on the improved neural network as claimed in claim 1, wherein: and the image enhancement method in the step two) comprises the steps of horizontal turning, vertical turning, 180-degree rotation, zooming, shifting, blurring, brightening, darkening and adding salt and pepper noise in sequence.
4. The method for identifying the tooth surface damage based on the improved neural network as claimed in claim 1, wherein: forming a classified tooth surface image data set in the second step, wherein the proportion of a training set to a testing set is 9:1.
5. the method for identifying the tooth surface damage based on the improved neural network as claimed in claim 1, wherein: the convolutional neural network in the third step sequentially comprises an input layer, a first convolutional layer, a second convolutional layer, a third convolutional layer, a fourth convolutional layer, a fifth convolutional layer, a global average pooling layer and a full-link layer.
6. The method for identifying tooth surface damage based on the improved neural network as claimed in claim 5, wherein the specific operation of constructing the convolutional neural network in the step three) is as follows:
1) The first winding layer
1.1 Passing the image with the dimension of (224, 64) through 64 convolution kernels of 7 by 7 and performing convolution operation with the step size of 2, and performing zero filling operation for 3 circles at the periphery of the obtained pixel matrix to obtain a three-dimensional characteristic map of (112, 64);
1.2 Carrying out batch normalization operation on the obtained three-dimensional characteristic diagram, and then carrying out nonlinear mapping operation on the batch normalized three-dimensional characteristic diagram through a ReLU activation function;
1.3 Performing maximal pooling operation on the feature map output in step 1.2) by 3-by-3 convolution kernel with 2 steps, and performing 1-turn zero-filling operation on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map (56, 64);
2) The second convolution layer
2.1 The three-dimensional characteristic diagram (56, 64) obtained in the previous step is convoluted by 64 convolution kernels of 3 x 3 and with the step size of 1, and the zero padding operation is carried out for 1 circle on the periphery of the obtained pixel matrix to obtain the three-dimensional characteristic diagram (56, 64);
2.2 Batch normalization operation is carried out on the three-dimensional feature maps (56, 64) obtained in the previous step, and then nonlinear mapping operation is carried out on the batch normalized three-dimensional feature maps through a ReLU activation function;
2.3 ) repeating the convolution and zero filling operation in the step 2.1) on the three-dimensional characteristic diagram obtained in the previous step, and carrying out batch normalization operation on the obtained three-dimensional characteristic diagram;
2.4 Adding the three-dimensional feature map input in the step 2.1) and the three-dimensional feature map output in the step 2.3), and then carrying out nonlinear mapping operation on the three-dimensional feature map obtained after addition through a ReLU activation function;
2.5 Repeating the step 2.1) to the step 2.4) twice, and outputting a three-dimensional feature map with dimensions (56, 64);
3) The third convolution layer
3.1 Carrying out convolution on the three-dimensional feature map with the dimension (56, 64) obtained in the previous step through 128 convolution kernels with 3 x 3 and the step size of 2, and carrying out 1-turn zero padding operation on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map (28, 128);
3.2 Carrying out batch normalization operation on the three-dimensional feature maps (28, 128) obtained in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature maps after batch normalization through a ReLU activation function to obtain three-dimensional feature maps with the dimension of (28, 64);
3.3 Carrying out convolution on the three-dimensional feature map with the dimension of (28, 64) obtained in the previous step through 128 convolution kernels of 3 x 3 and the step size of 1, carrying out zero filling operation for 1 circle on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map with the dimension of (28, 128), and carrying out batch normalization operation on the three-dimensional feature map;
3.4 Carrying out convolution on the three-dimensional feature map input in the step 3.1) through 128 convolution kernels of 1 × 1 and with the step size of 2 to obtain a three-dimensional feature map (28, 128), adding the three-dimensional feature map to the feature map obtained in the step 3.3), and then carrying out nonlinear mapping operation on the three-dimensional feature map obtained after addition through a ReLU activation function to obtain a three-dimensional feature map with the dimension of (28, 128);
3.5 Passing the three-dimensional feature map (28, 128) obtained in the previous step through 128 convolution kernels of 3 × 3 and performing convolution by step size of 1, and performing zero padding operation for 1 circle on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map (28, 128);
3.6 Carrying out batch normalization operation on the three-dimensional feature maps output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature maps subjected to batch normalization through a ReLU activation function;
3.7 Step 3.5) is repeated and batch normalization operation is carried out on the obtained three-dimensional characteristic diagram;
3.8 Adding the three-dimensional feature map output in the previous step with the three-dimensional feature map with the dimension (28, 128) input in the step 3.5), and then carrying out nonlinear mapping operation on the three-dimensional feature map obtained after addition through a ReLU activation function;
3.9 Repeating the step 3.5) to the step 3.8) twice, and outputting a three-dimensional feature map with the dimension of (28, 128);
4) The fourth convolution layer
4.1 Passing the three-dimensional feature map obtained in the previous step through 256 convolution kernels of 3 × 3 and performing convolution with 2 steps, and performing 1-turn zero padding operation on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map of (14, 256);
4.2 Carrying out batch normalization operation on the three-dimensional feature map (14, 256) obtained in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature map after batch normalization through a ReLU activation function to obtain a three-dimensional feature map with the dimension of (14, 256);
4.3 Passing the three-dimensional feature map obtained in the previous step through 256 convolution kernels of 3 x 3 and performing convolution by step size of 1, performing zero padding operation for 1 circle on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map (14, 256), and performing batch normalization operation on the three-dimensional feature map;
4.4 Carrying out convolution on the three-dimensional feature map input in the step 4.1) through 256 convolution kernels of 1 × 1 and with the step size of 2 to obtain a three-dimensional feature map with the dimension (14, 256) which is added to the three-dimensional feature map output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature map obtained after addition through a ReLU activation function;
4.5 Passing the three-dimensional feature map output in the previous step through 256 convolution kernels of 3 x 3 and performing convolution by step size of 1, and performing zero filling operation for 1 circle on the periphery of the obtained pixel matrix to obtain a three-dimensional feature map (14, 256);
4.6 Carrying out batch normalization operation on the three-dimensional characteristic diagram output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional characteristic diagram after batch normalization through a ReLU activation function;
4.7 Step 4.5) is repeated and the obtained three-dimensional characteristic diagram is subjected to batch normalization operation;
4.8 Adding the three-dimensional feature map input in the step 4.5) with the three-dimensional feature map output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature map obtained after addition through a ReLU activation function;
4.9 Repeating the step 4.5) to the step 4.8) four times, and outputting a three-dimensional feature map with the dimension of (14, 256);
5) The fifth convolution layer
5.1 Carrying out convolution on the three-dimensional feature map obtained in the previous step by 512 convolution kernels of 3 x 3 and 2 steps, and carrying out 1-turn zero filling operation on the periphery of the obtained pixel matrix to obtain a (7, 512) three-dimensional feature map;
5.2 Carrying out batch normalization operation on the three-dimensional characteristic diagram obtained in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional characteristic diagram after batch normalization through a ReLU activation function to obtain a three-dimensional characteristic diagram with the dimensionality of (7, 512);
5.3 The three-dimensional feature map obtained in the previous step is convoluted by 512 convolution kernels of 3 x 3 and the step length of 1, and the zero padding operation is carried out for 1 circle on the periphery of the obtained pixel matrix to obtain the three-dimensional feature map of (7, 512), and the three-dimensional feature map is subjected to batch normalization operation;
5.4 Carrying out convolution on the three-dimensional feature map input in the step 5.1) through 512 convolution kernels of 1 × 1 and 2 steps to obtain a three-dimensional feature map with the dimension of (7, 512), adding the three-dimensional feature map with the dimension of (7, 512) with the three-dimensional feature map output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature map obtained after addition through a ReLU activation function;
5.5 Carrying out convolution on the three-dimensional feature map output in the previous step by 512 convolution kernels of 3 x 3 and step length of 1, and carrying out 1-turn zero filling operation on the periphery of the obtained pixel matrix to obtain a (7, 512) three-dimensional feature map;
5.6 Carrying out batch normalization operation on the three-dimensional feature maps output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional feature maps subjected to batch normalization through a ReLU activation function;
5.7 Step 5.5) is repeated and batch normalization operation is carried out on the obtained three-dimensional characteristic diagram;
5.8 Adding the three-dimensional characteristic diagram input in the step 5.5) with the three-dimensional characteristic diagram output in the previous step, and then carrying out nonlinear mapping operation on the three-dimensional characteristic diagram obtained after addition through a ReLU activation function;
5.9 Repeating the step 5.5) to the step 5.8) once, and outputting a three-dimensional feature map with the dimension of (7, 512);
6) Global average pooling layer
Performing global average pooling operation on the three-dimensional feature map with the dimension of (7, 512) output in the previous step, and outputting the three-dimensional feature map with the dimension of (1, 512);
7) Full connection layer
And converting the three-dimensional feature map with the dimension of (1, 512) output in the previous step into a 4-dimensional feature vector.
7. The method for tooth surface damage identification based on the improved neural network as claimed in claim 6, wherein: after the batch normalization operation in the step 1.2), the step 2.2), the step 2.3), the step 3.2), the step 3.3), the step 3.6), the step 3.7), the step 4.2), the step 4.3), the step 4.6), the step 4.7), the step 5.2), the step 5.3), the step 5.6) and the step 5.7), pruning with a pruning rate of 20% is performed on each batch normalization layer.
8. The method for identifying tooth surface damage based on the improved neural network as claimed in claim 1, wherein the specific operations of training the constructed convolutional neural network and saving the formed weights into the constructed convolutional neural network are as follows:
a) Inputting a training set in the tooth surface image data set into a built convolutional neural network for training, and performing transfer learning by using official weight to finally form a plurality of output numerical values corresponding to the classification quantity;
b) Make itBy using
A function, converting a plurality of output values into relative probabilities to obtain probability vectors corresponding to each classification, and selecting the classification with the highest probability as a predicted value of the output;
where i is the index of the class, C is the number of classes, V i The output of a preceding stage output unit of the classifier of the full connection layer is shown, and e is a natural base number;
c) Inputting the obtained predicted value and the real label into a cross entropy function CrossEntropyLoss to calculate an error value, calculating a gradient value of each parameter through an error back propagation function loss.
d) Inputting the test set into a convolutional neural network to verify the training effect, calculating an error value according to a cross entropy function, outputting the type with the maximum probability as a predicted value, judging that the prediction is correct if the predicted value is the same as the real label, and determining the ratio of the number of correct predictions to the number of test sets as the accuracy of the whole test set;
e) Training for 400 times, wherein the lowest error value on the set to be tested is 0.095, the accuracy rate reaches 97.7%, and the trained weight is stored in the convolutional neural network.
9. An electronic device comprising a processor and a memory, the memory storing computer readable instructions, the computer readable instructions, when executed by the processor, being capable of performing the tooth surface damage identification method of any one of claims 1-8.
10. A readable storage medium, on which a computer program is stored, which, when being executed by a processor, performs the steps of the method for tooth flank damage identification according to any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211414681.7A CN115565020A (en) | 2022-11-11 | 2022-11-11 | Tooth surface damage identification method and device based on improved neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211414681.7A CN115565020A (en) | 2022-11-11 | 2022-11-11 | Tooth surface damage identification method and device based on improved neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115565020A true CN115565020A (en) | 2023-01-03 |
Family
ID=84770434
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211414681.7A Pending CN115565020A (en) | 2022-11-11 | 2022-11-11 | Tooth surface damage identification method and device based on improved neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115565020A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116091445A (en) * | 2023-01-05 | 2023-05-09 | 中国长江电力股份有限公司 | Ship lift gear and gear surface damage identification method based on deep learning |
| CN117636057A (en) * | 2023-12-13 | 2024-03-01 | 石家庄铁道大学 | Train bearing damage classification and identification method based on multi-branch cross-space attention model |
-
2022
- 2022-11-11 CN CN202211414681.7A patent/CN115565020A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116091445A (en) * | 2023-01-05 | 2023-05-09 | 中国长江电力股份有限公司 | Ship lift gear and gear surface damage identification method based on deep learning |
| CN116091445B (en) * | 2023-01-05 | 2024-01-02 | 中国长江电力股份有限公司 | Ship lift gear and gear surface damage identification method based on deep learning |
| CN117636057A (en) * | 2023-12-13 | 2024-03-01 | 石家庄铁道大学 | Train bearing damage classification and identification method based on multi-branch cross-space attention model |
| CN117636057B (en) * | 2023-12-13 | 2024-06-11 | 石家庄铁道大学 | Train bearing damage classification and identification method based on multi-branch cross-space attention model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115565020A (en) | Tooth surface damage identification method and device based on improved neural network | |
| CN110969175B (en) | Wafer processing method and device, storage medium and electronic equipment | |
| CN112561910A (en) | Industrial surface defect detection method based on multi-scale feature fusion | |
| CN116258707A (en) | PCB surface defect detection method based on improved YOLOv5 algorithm | |
| CN110751195B (en) | Fine-grained image classification method based on improved YOLOv3 | |
| CN114140683A (en) | Aerial image target detection method, equipment and medium | |
| CN113095370A (en) | Image recognition method and device, electronic equipment and storage medium | |
| CN114743189A (en) | Pointer instrument reading identification method and device, electronic equipment and storage medium | |
| TWI812888B (en) | Image recognition method and image recognition system | |
| CN112364974A (en) | Improved YOLOv3 algorithm based on activation function | |
| CN114971375A (en) | Examination data processing method, device, equipment and medium based on artificial intelligence | |
| CN114048817A (en) | Deep learning input set priority test method based on variation strategy | |
| CN116630700A (en) | Remote sensing image classification method based on introduction channel-space attention mechanism | |
| CN112381177A (en) | Dial digital character recognition method and system based on deep learning | |
| CN110321867B (en) | Shielded target detection method based on component constraint network | |
| CN113763364B (en) | Image defect detection method based on convolutional neural network | |
| US11195265B2 (en) | Server and method for recognizing image using deep learning | |
| CN113541985A (en) | Internet of things fault diagnosis method, training method of model and related device | |
| CN105654138A (en) | Orthogonal projection and dimensionality reduction classification method and system for multidimensional data | |
| CN116363136B (en) | On-line screening method and system for automatic production of motor vehicle parts | |
| CN113421223A (en) | Industrial product surface defect detection method based on deep learning and Gaussian mixture | |
| CN111666872A (en) | Efficient behavior identification method under data imbalance | |
| CN114743023A (en) | Wheat spider image detection method based on RetinaNet model | |
| CN114861771A (en) | Industrial CT image defect classification method based on feature extraction and deep learning | |
| CN113971737A (en) | Object recognition method for robot, electronic device, medium, and program product |
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 | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20240703 Address after: 450001 No.149, science Avenue, high tech Zone, Zhengzhou City, Henan Province Applicant after: Zheng Ji Suo (Zhengzhou) Transmission Technology Co.,Ltd. Country or region after: China Address before: 450001 149 science Avenue, Zhengzhou high tech Industrial Development Zone, Henan Applicant before: ZHENGZHOU RESEARCH INSTITUTE OF MECHANICAL ENGINEERING Co.,Ltd. Country or region before: China |
|
| TA01 | Transfer of patent application right |