Object detection method and device, storage medium, controller and program product

Abstract

The invention relates to an object detection method and device, a storage medium, a controller and a program product, and the method comprises the steps: determining an ultrasonic signal and a pulsed eddy current signal of a to-be-detected object; extracting a first characteristic signal in the ultrasonic signal and a second characteristic signal in the pulsed eddy current signal; and inputting the first feature signal and the second feature signal into a first convolutional neural network model to obtain a defect identification result and a thickness identification result of the to-be-detected object, the first convolutional neural network model being used for identifying whether the to-be-detected object has defects and whether the thickness value is within a preset thickness threshold range. The first feature signal and the second feature signal are subjected to data fusion through the first convolutional neural network model, so that the defect recognition result and the thickness recognition result can be obtained, and compared with the prior art in which defect detection and thickness detection are separately detected or analyzed, the calculation amount can be reduced, and the detection efficiency can be improved. And meanwhile, the accuracy and efficiency of a detection result can be improved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of signal processing, in particular, to an object detection method and device, a storage medium, a controller and a program product. BACKGROUND

[0002] In the related art, when there is a defect or damage on the surface of an object, a high-precision flaw detection method is usually used to detect whether there is a defect on the surface of the object. When the thickness of the object needs to be measured, a thickness measurement method is usually used to detect the thickness of the object. SUMMARY

[0003] The purpose of the present disclosure is to provide an object detection method, device, storage medium, controller and program product to solve the technical problems existing in the related art.

[0004] To achieve the above purpose, in a first aspect, the present disclosure provides an object detection method, comprising: determining an ultrasonic signal and a pulse eddy current signal of a to-be-detected object; extracting a first feature signal in the ultrasonic signal and a second feature signal in the pulse eddy current signal; inputting the first feature signal and the second feature signal into a first convolutional neural network model to obtain a defect identification result and a thickness identification result of the to-be-detected object, wherein the first convolutional neural network model is used to identify whether the to-be-detected object has a defect and whether the thickness value is within a preset thickness threshold range.

[0005] Optionally, the inputting the first feature signal and the second feature signal into the first convolutional neural network model to obtain the defect identification result and the thickness identification result of the to-be-detected object comprises: sequentially passing the first feature signal and the second feature signal through a first convolutional layer, a first pooling layer, a second pooling layer, a connection layer, a second convolutional layer, a third pooling layer, a fourth pooling layer, a first binary classifier and a second binary classifier in the first convolutional neural network model to obtain the defect identification result and the thickness identification result; wherein the first binary classifier is used to identify whether the to-be-detected object has a defect, and the second binary classifier is used to identify whether the thickness value of the to-be-detected object is within a preset thickness threshold range.

[0006] Optionally, when the defect identification result is used to represent that the surface of the to-be-detected object has a defect, inputting the defect identification result into a second convolutional neural network model to obtain defect information of the to-be-detected object, wherein the second convolutional neural network model is used to identify the defect information of the to-be-detected object.

[0007] Optionally, the inputting the defect identification result into the second convolutional neural network model to obtain the defect information of the object to be detected comprises: extracting defect feature data in the defect identification result; extracting defect position information in the defect feature data according to a region prediction network in the second convolutional neural network model, performing full convolution processing on the defect position information, classifying the pooling features of the defect position information after the full convolution processing to obtain weights, and performing convolution processing on the weights and preset feature quantities to obtain defect label semantics; fusing the defect position information and the defect label semantics, and performing convolution on the fused defect position information and defect label semantics to obtain label semantics of the object to be detected; predicting the label semantics of the object to be detected to obtain a predicted defect category; determining the defect position information and the predicted defect category as the defect information.

[0008] Optionally, the extracting the first feature signal in the ultrasonic signal and the second feature signal in the pulse eddy current signal comprises: extracting the first feature signal in the ultrasonic signal and the second feature signal in the pulse eddy current signal through a self-encoder, wherein the self-encoder is a self-encoder with a preset threshold number of hidden layers and a sparse penalty function.

[0009] In a second aspect, the present disclosure provides an object detection device, comprising a controller, a first electromagnetic ultrasonic magnet, a first electromagnetic ultrasonic coil, a second electromagnetic ultrasonic magnet, a second electromagnetic ultrasonic coil, a third electromagnetic ultrasonic coil, a first pulse eddy current coil, a second pulse eddy current coil, and a ferrite shielding layer. The controller is connected with the first electromagnetic ultrasonic coil, the second electromagnetic ultrasonic coil, the third electromagnetic ultrasonic coil, the first pulse eddy current coil, and the second pulse eddy current coil, respectively. The first pulse eddy current coil is arranged between the first electromagnetic ultrasonic magnet and the third electromagnetic ultrasonic coil. The second pulse eddy current coil is arranged between the third electromagnetic ultrasonic coil and the second electromagnetic ultrasonic magnet. The first electromagnetic ultrasonic coil is arranged below the first electromagnetic ultrasonic magnet. The second electromagnetic ultrasonic coil is arranged below the second electromagnetic ultrasonic magnet. The third electromagnetic ultrasonic coil is arranged in the ferrite shielding layer. The controller is configured to control the first and second electromagnetic ultrasonic coils to emit detection signals to a to-be-detected object, receive ultrasonic signals transmitted by the third electromagnetic ultrasonic coil, and control the first pulsed eddy current coil to emit detection signals to the to-be-detected object, receive the pulsed eddy current signals input by the second pulsed eddy current coil, or control the second pulsed eddy current coil to emit detection signals to the to-be-detected object, receive the pulsed eddy current signals input by the first pulsed eddy current coil, so as to implement the object detection method according to any one of the first aspect of the present disclosure.

[0010] In a third aspect, the present disclosure provides an object detection device, comprising a first determination module, a signal extraction module, and an identification module. The first determination module is configured to determine ultrasonic signals and pulsed eddy current signals of a to-be-detected object. The signal extraction module is configured to extract first characteristic signals in the ultrasonic signals and second characteristic signals in the pulsed eddy current signals. The identification module is configured to input the first characteristic signals and the second characteristic signals into a first convolutional neural network model to obtain a defect identification result and a thickness identification result of the to-be-detected object, wherein the first convolutional neural network model is configured to identify whether the to-be-detected object has a defect and whether a thickness value of the to-be-detected object is within a preset thickness threshold range.

[0011] Optionally, the identification module is configured to: pass the first characteristic signals and the second characteristic signals through a first convolutional layer, a first pooling layer, a second pooling layer, a connection layer, a second convolutional layer, a third pooling layer, a fourth pooling layer, a first binary classifier, and a second binary classifier in the first convolutional neural network model in sequence to obtain the defect identification result and the thickness identification result. The first binary classifier is configured to identify whether the to-be-detected object has a defect, and the second binary classifier is configured to identify whether the thickness value of the to-be-detected object is within the preset thickness threshold range.

[0012] Optionally, when the defect identification result is used to represent that the to-be-detected object has a defect on the surface, the defect identification result is input into a second convolutional neural network model to obtain defect information of the to-be-detected object, wherein the second convolutional neural network model is configured to identify the defect information of the to-be-detected object.

[0013] Optionally, the inputting of the defect identification result into the second convolutional neural network model to obtain the defect information of the to-be-detected object comprises: extracting defect feature data in the defect identification result. According to the region prediction network in the second convolutional neural network model, defect position information in the defect feature data is extracted, the defect position information is fully convolutional processed, and the pooled features of the defect position information after the full convolutional processing are classified to obtain weights. The weights are convolutional processed with preset feature quantities to obtain defect label semantics; The defect position information and the defect label semantics are fused, and the defect position information and the defect label semantics after the fusion are convolutional processed to obtain label semantics of the object to be tested; The label semantics of the object to be tested is predicted to obtain a predicted defect category; The defect position information and the predicted defect category are determined as the defect information.

[0014] Optionally, the signal extraction module is configured to: extract the first feature signal in the ultrasonic signal and the second feature signal in the pulsed eddy current signal through a self-encoder, wherein the self-encoder is a self-encoder with a preset threshold number of hidden layers and a sparse penalty function.

[0015] In a fourth aspect, the present disclosure provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the steps of any of the methods provided in the first aspect of the present disclosure.

[0016] In a fifth aspect, the present disclosure provides a controller, comprising: a memory having a computer program stored thereon; a processor configured to execute the computer program in the memory to implement the steps of any of the methods provided in the first aspect of the present disclosure.

[0017] In a sixth aspect, the present disclosure provides a computer program product comprising a computer program, wherein the computer program is executed by a processor to implement the steps of any of the methods provided in the first aspect of the present disclosure.

[0018] By the technical solution, the first characteristic signal in the ultrasonic signal of the to-be-detected object and the second characteristic signal in the pulse eddy current signal are input into the first convolutional neural network model, and the first characteristic signal and the second characteristic signal are processed by the first convolutional neural network model to obtain the defect recognition result and the thickness recognition result of the to-be-detected object. The first characteristic signal and the second characteristic signal are input into the first convolutional neural network model for analysis by the first convolutional neural network model, so that the first characteristic signal and the second characteristic signal can be data fused by the first convolutional neural network model, and then the defect recognition result and the thickness recognition result can be obtained. Compared with the defect detection and thickness detection being separately detected or analyzed in the related art, the calculation amount can be reduced, and the accuracy and efficiency of the detection result can be improved.

[0019] Other features and advantages of the present disclosure will be described in detail in the following detailed description section. BRIEF DESCRIPTION OF DRAWINGS

[0020] The accompanying drawings are included to provide a further understanding of the present disclosure and constitute a part of the specification, and are used together with the following detailed description to explain the present disclosure, but do not constitute a limitation of the present disclosure. In the drawings: Figure 1 FIG. 1 is a schematic diagram illustrating an object detection method according to an example embodiment of the present disclosure.

[0021] Figure 2 FIG. 2 is a schematic diagram illustrating a first characteristic signal and a second characteristic signal according to an example embodiment of the present disclosure.

[0022] Figure 3 FIG. 3 is a schematic diagram illustrating a self-encoder according to an example embodiment of the present disclosure.

[0023] Figure 4 FIG. 4 is a schematic diagram illustrating a first convolutional neural network model according to an example embodiment of the present disclosure.

[0024] Figure 5 FIG. 5 is a schematic diagram illustrating a second convolutional neural network model according to an example embodiment of the present disclosure.

[0025] Figure 6 FIG. 6 is a schematic diagram illustrating an object detection device according to an example embodiment of the present disclosure.

[0026] Figure 7 FIG. 7 is a schematic diagram illustrating an object detection device according to an example embodiment of the present disclosure.

[0027] REFERENCE SIGNS 1, first electromagnetic ultrasonic magnet; 2, first electromagnetic ultrasonic coil; 3, first pulse eddy current coil; 4, ferrite shielding layer; 5, third electromagnetic ultrasonic coil; 6, second pulse eddy current coil; 7, second electromagnetic ultrasonic magnet; 8, second electromagnetic ultrasonic coil; 9, object to be measured; 700, object detection device; 701, first determination module; 702, signal extraction module; 703, identification module. DETAILED DESCRIPTION

[0028] The specific embodiments of the present disclosure are described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present disclosure, and are not used to limit the present disclosure.

[0029] It should be noted that all actions of obtaining signals, information or data in the present disclosure are carried out in accordance with the corresponding data protection regulations and policies of the country where the device is located, and with the authorization given by the corresponding device owner.

[0030] High-precision flaw detection and thickness measurement are crucial for ensuring the safety of equipment detection. Many damages of large equipment cannot be detected by the naked eye, and can be detected by Pulsed Eddy Current Testing (PECT). At the same time, electromagnetic acoustic transducer (EMAT) is used for thickness measurement. The damage degree of the equipment can be confirmed by combining the two measurement results. When the equipment is a special working environment equipment and the surface condition of the equipment is complex, although the conventional ultrasonic has the advantages of simple principle and convenient implementation, it cannot meet the thickness measurement standard in this special situation, and even is not suitable for some situations. The commonly used flaw detection techniques include ultrasonic detection, magnetic powder detection, and penetration detection. Considering the working state of the equipment, the pipeline is filled with water, so these methods are not suitable.

[0031] Ultrasonic waves are sound waves with a frequency exceeding 20KHz, which have strong penetration and energy concentration. In today's social production and life, electromagnetic ultrasonic thickness measurement, as a widely used non-contact ranging method, plays a crucial role in industrial measurement due to its low cost, simple structure, and strong applicability.

[0032] The transmission circuit of electromagnetic ultrasonic transmits high-voltage high-frequency signals into the detection coil, which makes the crystal inside the object to be measured vibrate under the action of the permanent magnet to produce ultrasonic waves. When the detection coil detects the signal again, the received circuit amplifies, filters, and reduces noise to eliminate noise and improve the signal-to-noise ratio. After analyzing the echo signal characteristics, the echo information is obtained.

[0033] In pulsed eddy current nondestructive testing, an excitation square wave signal with a specific duty cycle is applied to the excitation coil, generating a primary magnetic field around the coil. This primary magnetic field approaches the test piece, inducing an exponentially decaying eddy current on the test piece surface along the longitudinal direction. This changing eddy current, in turn, generates a secondary magnetic field that acts in the opposite direction to the primary magnetic field. When the properties of the test piece change, such as defects, conductivity, or permeability, the induced eddy current density within the test piece will change. The detection coil picks up these eddy current changes and converts them into an induced voltage signal, which is then used to characterize defect information in the test piece.

[0034] In the first related technology, a non-destructive testing method for bulk defects based on a combination of pulsed eddy current and electromagnetic ultrasound is used to detect the object under test. This method combines the advantages of pulsed eddy current and electromagnetic ultrasound to achieve quantitative non-destructive testing of surface defects, thickness, and depth defects of conductive plate materials.

[0035] In the second related technology, the input signals of electromagnetic ultrasound and pulsed eddy current are loaded together and introduced into the detection coil. After the detection is completed, the obtained signals are separated.

[0036] In the third related technology, when performing thickness measurement and flaw detection on an object, the two tasks are usually performed separately, and the measurement results are analyzed separately.

[0037] The inventors discovered that when detecting an object using the first correlation technique, the ability of pulsed eddy current signals to detect deep defects decreases, while ultrasound is ineffective for detecting surface and near-surface defects. This leads to reduced accuracy of the detection results. When detecting an object using the second correlation technique, signal interference occurs between the ultrasonic and pulsed eddy current signals during measurement, resulting in inaccurate detection results. When detecting an object using the third correlation technique, the computational load is high, and inaccurate results may occur if the object surface is very rough or uneven.

[0038] In view of this, the present disclosure provides object detection methods, apparatus, storage media, controllers and program products to solve the technical problems existing in the related art.

[0039] like Figure 1 As shown, Figure 1 This is a schematic diagram illustrating an object detection method according to an exemplary embodiment of the present disclosure, with reference to... Figure 1 ,include: S101: Determine the ultrasonic signal and pulsed eddy current signal of the object under test; S102: Extract the first feature signal from the ultrasonic signal and the second feature signal from the pulsed eddy current signal; S103: Input the first feature signal and the second feature signal into the first convolutional neural network model to obtain the defect identification result and thickness identification result of the object under test, wherein the first convolutional neural network model is used to identify whether the object under test has defects and whether the thickness value is within the preset thickness threshold range.

[0040] The above technical solution involves inputting the first feature signal from the ultrasonic signal detected in the object under test and the second feature signal from the pulsed eddy current signal into a first convolutional neural network model. The first convolutional neural network model then processes these signals to obtain the defect identification result and the thickness identification result of the object under test. By inputting the first and second feature signals into the first convolutional neural network model for analysis, data fusion of the first and second feature signals can be achieved, thereby obtaining the defect identification result and the thickness identification result. Compared to related technologies that separate defect detection and thickness detection for detection or analysis, this reduces the computational load and improves the accuracy and efficiency of the detection results.

[0041] To enable those skilled in the art to better understand the object detection method provided in this disclosure, the above steps are illustrated in detail below.

[0042] For example, an ultrasonic signal can be an electromagnetic ultrasonic signal, which can be a signal used to detect internal defects, structures, or properties of an object under test through the principle of electromagnetic induction. This electromagnetic ultrasonic signal can be obtained by measuring with an electromagnetic ultrasonic magnet. A pulsed eddy current signal can be a signal generated by rapidly changing the current. This pulsed eddy current signal can be obtained by measuring with a pulsed eddy current coil.

[0043] For example, such as Figure 2 As shown, the first characteristic signal can be the peak signal, the peak time, and the rise time of the ultrasonic signal in the ultrasonic signal; the second characteristic signal includes the voltage of the pulse eddy current coil and the resistance of the pulse eddy current coil in the pulse eddy current signal. In this embodiment, no specific limitation is made.

[0044] In this embodiment, a first feature signal of the ultrasonic signal and a second feature signal of the pulsed eddy current signal can be extracted. The first feature signal and the second feature signal can be used in subsequent steps to perform data fusion analysis on the first feature signal and the second feature signal to obtain the defect identification result and thickness identification result of the object under test.

[0045] In a possible manner, the extraction of the first feature signal from the ultrasound signal and the second feature signal from the pulsed eddy current signal includes: extracting the first feature signal in the ultrasonic signal and the second feature signal in the pulsed eddy current signal through a self-encoder, wherein the self-encoder is a self-encoder with a preset threshold number of layers of hidden layers and a sparse penalty function.

[0046] It should be understood that the self-encoder can belong to a neural network and can be used to extract a data feature signal. Figure 3 As shown in the self-encoder includes an encoding network and a decoding network, Figure 3 The encoding network ~ The decoding network respectively realizes the functions of importing original data and exporting feature quantities, wherein the number of the imported original data is equal to the number of the exported feature quantities. The self-encoder can be used to reconstruct the input, taking the output of the previous layer as the input of the next layer, so as to reduce the difference between the input and the output and achieve the purpose of extracting features. The original data can be the ultrasonic signal and the pulsed eddy current signal, and the feature quantities can be the first feature signal and the second feature signal, which are not limited in the embodiments of the present disclosure. Therefore, the first feature signal in the ultrasonic signal and the second feature signal in the pulsed eddy current signal can be extracted through the self-encoder.

[0047] In the feature quantities extracted by the self-encoder, there are many factors that affect the quality of the feature quantities, such as the number of hidden layers, the number of nodes in each layer, the type of activation function in the hidden layer, the learning rate, the vector dimension, etc., but the most influential factor is the number of hidden layers. When the number of encoder layers is too small, the feature quantities extracted by the self-encoder do not fit the original data well; when the number of encoder hidden layers is too large, overfitting occurs, resulting in that the extracted feature quantities cannot well reflect the most essential features of the original data. Therefore, the preset threshold number of layers can be set to 4 layers, and the sparse penalty function can be set to , which is not limited in the embodiments of the present disclosure. Setting the preset threshold number of layers to 4 layers can extract the first feature signal and the second feature signal with better quality. Setting the sparse penalty function to , which can reduce the redundant feature quantities in the extracted first feature signal and the second feature signal.

[0048] The encoding network can be expressed by the following calculation formula:

[0049]

[0050] , wherein, is the self-encoder, is the first layer output of the self-encoder, ​a weight of a first layer of the autoencoder, a correction value of the first layer of the autoencoder, an output of a second layer of the autoencoder, a weight of the second layer of the autoencoder, a correction value of the second layer of the autoencoder, a number of first layers arranged in the autoencoder, a number of second layers arranged in the autoencoder.

[0051] The decoding network can be calculated by the following calculation formula.

[0052]

[0053] wherein, an output of the decoding network, a weight in the decoding network, a correction value of the decoding network, a number of decoding networks arranged in the autoencoder.

[0054] By way of example, a convolutional neural network can be used for processing images and videos, and its core idea can achieve image recognition and classification by learning the features of images. The convolutional neural network is composed of multiple convolutional layers, pooling layers and fully connected layers, which work together to complete feature extraction and classification of images. In the convolutional layer, the input image can be convolved by using a certain convolution kernel to extract features. In the pooling layer, the dimension of the input image can be reduced by using the pooling operation to reduce the number of parameters and improve the generalization ability of the model. In the fully connected layer, the input features can be classified by using the full connection operation.

[0055] The first convolutional neural network model can be a convolutional neural network model obtained by training historical detection data. The historical detection data can be historical ultrasonic signals and corresponding defect recognition results, historical pulsed eddy current signals and corresponding thickness recognition results, or the historical detection data can be ultrasonic signals and corresponding defect recognition results, and pulsed eddy current signals and corresponding thickness recognition results set by relevant personnel. Then the first convolutional neural network model can be used to identify whether the object to be measured has a defect and whether the thickness value of the object to be measured is within a preset thickness threshold range.

[0056] In this embodiment, the first feature signal and the second feature signal can be simultaneously input into the first convolutional neural network model, and the first convolutional neural network model can be used to process the first feature signal and the second feature signal to obtain the defect recognition result and the thickness recognition result.

[0057] By inputting the first feature signal and the second feature signal into the first convolutional neural network model, the first convolutional neural network model can perform data fusion on the first feature signal and the second feature signal, and simultaneously obtain the defect identification result and the thickness identification result. Compared with the related art in which defect detection and thickness detection are separately detected or analyzed, the amount of calculation can be reduced, and the accuracy and efficiency of the detection result can be improved.

[0058] In a possible manner, the inputting the first feature signal and the second feature signal into the first convolutional neural network model to obtain the defect identification result and the thickness identification result of the object to be detected includes: The first feature signal and the second feature signal sequentially pass through a first convolutional layer, a first pooling layer, a second pooling layer, a connection layer, a second convolutional layer, a third pooling layer, a fourth pooling layer, a first binary classifier, and a second binary classifier in the first convolutional neural network model to obtain the defect identification result and the thickness identification result. The first binary classifier is configured to identify whether the object to be detected has a defect, and the second binary classifier is configured to identify whether the thickness value of the object to be detected is within a preset thickness threshold range.

[0059] It should be understood that, in the first convolutional neural network model, as shown in FIG. 1, the first convolutional layer, the first pooling layer, the second pooling layer, the connection layer, the second convolutional layer, the third pooling layer, the fourth pooling layer, the first binary classifier, and the second binary classifier can be arranged. Figure 4 The arrangement of the first pooling layer, the second pooling layer, the third pooling layer, and the fourth pooling layer can reduce model overfitting and reduce the training time of the first convolutional neural network model. The four layers of maximum pooling layers can reduce the number of parameters to be learned, thereby reducing the calculation cost and providing basic translation invariance for internal features. The arrangement of the first binary classifier and the second binary classifier can determine the defect identification result of the object to be detected through the first binary classifier and determine the thickness identification result of the object to be detected through the second binary classifier. The defect identification result can indicate that the object to be detected has a defect or that the object to be detected does not have a defect, and the thickness identification result can be a specific thickness value. The first convolutional neural network model can then fuse the first feature signal and the second feature signal to obtain the defect identification result and the thickness identification result.

[0060] In a possible manner, when the defect identification result indicates that the surface of the object to be detected has a defect, the defect identification result is input into a second convolutional neural network model to obtain defect information of the object to be detected. The second convolutional neural network model is configured to identify the defect information of the object to be detected.

[0061] It should be understood that when the defect identification result represents that the surface of the to-be-tested object has defects, the defect identification result can be further processed by the second convolutional neural network model to obtain defect information of the to-be-tested object. The defect information can include defect position information and a predicted defect category, and the embodiments of the present disclosure do not make specific limitations thereto. The second convolutional neural network model can be a convolutional neural network model obtained by training historical data. The historical data can be defect information data obtained by historical detection.

[0062] In a possible manner, the inputting of the defect identification result into the second convolutional neural network model to obtain the defect information of the to-be-tested object includes: extracting defect feature data in the defect identification result; extracting defect position information in the defect feature data according to a region prediction network in the second convolutional neural network model, performing full convolution processing on the defect position information, classifying pooled features of the defect position information after the full convolution processing to obtain a weight, and performing convolution processing on the weight and a preset feature quantity to obtain defect label semantics; fusing the defect position information and the defect label semantics, and performing convolution on the fused defect position information and defect label semantics to obtain label semantics of the to-be-tested object; predicting the label semantics of the to-be-tested object to obtain a predicted defect category; determining the defect position information and the predicted defect category as the defect information.

[0063] It should be understood that the defect information includes multiple different types of defects, and different defects correspond to different defect causes. The probabilities of different defects appearing in different positions are different. For example, near the welding of a steel plate, internal welds are more likely to appear, and on the surface of the steel plate, some surface defects are caused by long-term use. At the same time, since the positions of the defects can be related, it can be assumed that there is some correlation between the positions of the defects, and a correlation factor can be introduced into the second convolutional neural network model, and the local condition features of the steel plate and the overall condition are adaptively convolved.

[0064] As Figure 5As shown, the second convolutional neural network model can include a global feature extraction module, a defect category learning module, and an adaptive convolutional network module. The defect feature data in the defect recognition result can be extracted through the global feature extraction module. The global feature extraction model includes Conv (convolution), ROI (Region of Interest Pooling, region of interest), and Pooling (pooling), the defect category learning module includes FC (Fully Connected Layer, fully connected layer), RDPNblock (Residual Deep Pyramid Network Block, residual deep pyramid network block), and classifier (classifier). The adaptive convolutional network module includes AGCNblock (Attention-guided Graph Convolutional Network block, a module containing graph convolution operation and attention mechanism) and classifier. In the second convolutional neural network module, Backone (backbone network) is also included.

[0065] The defect location information in the defect feature data can be extracted by the defect category learning module according to the region prediction network in the second convolutional neural network model, the defect location information is fully convolutional processed, and the pooling features of the defect location information after the full convolutional processing are classified to obtain weights. The weights are convoluted with the preset feature quantity to obtain defect label semantics.

[0066] The defect location information and the defect label semantics can be fused by the adaptive convolutional network module, and the fused defect location information and the defect label semantics are convoluted to obtain the label semantics of the object to be tested. The label semantics of the object to be tested is predicted by a binary classifier to obtain a predicted defect category; and the defect location information and the predicted defect category are determined as the defect information.

[0067] By setting the second convolutional neural network module, the defect location information of the object to be tested and the predicted defect category can be determined. At the same time, the defect recognition result, the defect information, and the thickness recognition result can be classified and recognized, and the recognized result and the feature set extracted by the autoencoder are fused in features. The parameters are continuously updated in the end-to-end learning process to realize the label of the defect category of the convolutional network.

[0068] Based on the same concept, the embodiment also discloses an object detection device, which comprises a controller, a first electromagnetic ultrasonic magnet, a first electromagnetic ultrasonic coil, a second electromagnetic ultrasonic magnet, a second electromagnetic ultrasonic coil, a third electromagnetic ultrasonic coil, a first pulse eddy current coil, a second pulse eddy current coil, and a ferrite shielding layer. The controller is connected with the first electromagnetic ultrasonic coil, the second electromagnetic ultrasonic coil, the third electromagnetic ultrasonic coil, the first pulse eddy current coil and the second pulse eddy current coil respectively, the first pulse eddy current coil is arranged between the first electromagnetic ultrasonic magnet and the third electromagnetic ultrasonic coil, the second pulse eddy current coil is arranged between the third electromagnetic ultrasonic coil and the second electromagnetic ultrasonic magnet, the first electromagnetic ultrasonic coil is arranged below the first electromagnetic ultrasonic magnet, the second electromagnetic ultrasonic coil is arranged below the second electromagnetic ultrasonic magnet, and the third electromagnetic ultrasonic coil is arranged in the ferrite shielding layer. The controller is used for controlling the first electromagnetic ultrasonic coil and the second electromagnetic ultrasonic coil to emit detection signals to a to-be-detected object, receiving ultrasonic signals transmitted by the third electromagnetic ultrasonic coil, and controlling the first pulse eddy current coil to emit detection signals to the to-be-detected object, receiving pulse eddy current signals input by the second pulse eddy current coil, or controlling the second pulse eddy current coil to emit detection signals to the to-be-detected object, receiving pulse eddy current signals input by the first pulse eddy current coil, so as to realize the object detection method disclosed in the embodiment.

[0069] For example, as shown in the object detection device, Figure 6 The first electromagnetic ultrasonic coil 2 and the second electromagnetic ultrasonic coil 8 can emit detection signals to the to-be-detected object 9, and the third electromagnetic ultrasonic coil 5 can be used to receive ultrasonic signals after passing through the to-be-detected object 9 and input the ultrasonic signals into the controller. The third electromagnetic ultrasonic coil 5 is arranged in the ferrite shielding layer 4, the ferrite shielding layer 4 is arranged between the first pulse eddy current coil 3 and the second pulse eddy current coil 6, the first pulse eddy current coil 3 is arranged between the first electromagnetic ultrasonic magnet 1 and the third electromagnetic ultrasonic coil 5, and the second pulse eddy current coil 6 is arranged between the third electromagnetic ultrasonic coil 5 and the second electromagnetic ultrasonic magnet 7, so that the signal interference problems between the three electromagnetic ultrasonic coils and the two pulse eddy current coils can be avoided, and the ultrasonic signals and the pulse eddy current signals of the to-be-detected object can be detected at the same time.

[0070] In the first pulse eddy current coil 3 and the second pulse eddy current coil 6, the first pulse eddy current coil 3 can be used to emit detection signals to the to-be-detected object 9, and the second pulse eddy current coil 6 can be used to receive pulse eddy current signals and input the pulse eddy current signals into the controller. The second pulse eddy current coil 6 can also be used to emit detection signals to the to-be-detected object 9, and the first pulse eddy current coil 3 can be used to receive pulse eddy current signals and input the pulse eddy current signals into the controller.

[0071] The object detection device is arranged to simultaneously detect the ultrasonic signal and the pulse eddy current signal of the to-be-detected object, and input the ultrasonic signal and the pulse eddy current signal to the first convolutional neural network model for data fusion analysis and processing. The first characteristic signal and the second characteristic signal are fused by the first convolutional neural network model, and then the defect recognition result and the thickness recognition result can be obtained. Compared with the related art in which the defect detection and the thickness detection are separately detected or analyzed, the calculation amount can be reduced, and the accuracy and efficiency of the detection result can be improved.

[0072] Based on the same concept, the embodiment also discloses an object detection device 700, which is described with reference to Figure 7 , Figure 7 is a schematic diagram of an object detection device 700 according to an example embodiment of the present disclosure, as Figure 7 shown, comprising a first determination module 701, a signal extraction module 702, and an identification module 703; The first determination module 701 is configured to determine the ultrasonic signal and the pulse eddy current signal of the to-be-detected object. The signal extraction module 702 is configured to extract a first characteristic signal in the ultrasonic signal and a second characteristic signal in the pulse eddy current signal. The identification module 703 is configured to input the first characteristic signal and the second characteristic signal into a first convolutional neural network model to obtain a defect recognition result and a thickness recognition result of the to-be-detected object, wherein the first convolutional neural network model is configured to identify whether the to-be-detected object has a defect and whether the thickness value is within a preset thickness threshold range.

[0073] Optionally, the identification module 703 is configured to: pass the first characteristic signal and the second characteristic signal through a first convolutional layer, a first pooling layer, a second pooling layer, a connection layer, a second convolutional layer, a third pooling layer, a fourth pooling layer, a first binary classifier, and a second binary classifier in the first convolutional neural network model in sequence to obtain the defect recognition result and the thickness recognition result. The first binary classifier is configured to identify whether the to-be-detected object has a defect, and the second binary classifier is configured to identify whether the thickness value of the to-be-detected object is within a preset thickness threshold range.

[0074] Optionally, when the defect recognition result is used to represent that the to-be-detected object has a defect on the surface, the defect recognition result is input into a second convolutional neural network model to obtain defect information of the to-be-detected object, wherein the second convolutional neural network model is configured to identify the defect information of the to-be-detected object.

[0075] Optionally, the inputting the defect identification result into the second convolutional neural network model to obtain the defect information of the object to be detected comprises: extracting defect feature data in the defect identification result; extracting defect position information in the defect feature data according to a region prediction network in the second convolutional neural network model, performing full convolution processing on the defect position information, classifying pooled features of the defect position information after the full convolution processing to obtain weights, and performing convolution processing on the weights and preset feature quantities to obtain defect label semantics; fusing the defect position information and the defect label semantics, and performing convolution on the fused defect position information and defect label semantics to obtain label semantics of the object to be detected; predicting the label semantics of the object to be detected to obtain a predicted defect category; determining the defect position information and the predicted defect category as the defect information.

[0076] Optionally, the signal extraction module 702 is configured to: extract the first feature signal in the ultrasonic signal and the second feature signal in the pulsed eddy current signal through a self-encoder, wherein the self-encoder is a self-encoder with a preset threshold number of hidden layers and a sparse penalty function.

[0077] As to the apparatus in the above embodiments, the specific manners in which various modules perform operations have been described in detail in the embodiments of the method, and thus will not be described here in detail.

[0078] Based on the same idea, the present embodiment further discloses a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the object detection method disclosed in the present embodiment.

[0079] Based on the same idea, the present embodiment further discloses a controller, comprising: a memory having a computer program stored thereon; a processor configured to execute the computer program in the memory to implement the steps of the object detection method disclosed in the present embodiment.

[0080] In another exemplary embodiment, a computer program product is also provided, which contains a computer program executable by a programmable device, the computer program having code portions for performing the above-mentioned object detection method when executed by the programmable device.

[0081] The preferred embodiments of the present disclosure are described in detail above with reference to the drawings, but the present disclosure is not limited to the specific details in the above-described embodiments. Various simple modifications can be made to the technical solutions of the present disclosure within the technical concept of the present disclosure, and these simple modifications all belong to the protection scope of the present disclosure.

[0082] In addition, it should be noted that each specific technical feature described in the above specific embodiments can be combined in any appropriate manner without contradiction. In order to avoid unnecessary repetition, various possible combinations are not described again in the present disclosure.

[0083] In addition, various different embodiments of the present disclosure can also be combined in any appropriate manner, as long as they do not deviate from the idea of the present disclosure, and they should also be considered as disclosed in the present disclosure.

Claims

1. An object detection method characterized by, The method comprises the following steps: determining an ultrasonic signal and a pulse eddy current signal of a to-be-tested object; extracting a first feature signal in the ultrasonic signal and a second feature signal in the pulse eddy current signal; inputting the first feature signal and the second feature signal into a first convolutional neural network model to obtain a defect recognition result and a thickness recognition result of the to-be-tested object, wherein the first convolutional neural network model is used to identify whether the to-be-tested object has a defect and whether a thickness value is within a preset thickness threshold range.

2. The object detection method according to claim 1, wherein The inputting the first feature signal and the second feature signal into the first convolutional neural network model to obtain the defect recognition result and the thickness recognition result comprises: sequentially inputting the first feature signal and the second feature signal through a first convolutional layer, a first pooling layer, a second pooling layer, a connection layer, a second convolutional layer, a third pooling layer, a fourth pooling layer, a first binary classifier and a second binary classifier in the first convolutional neural network model to obtain the defect recognition result and the thickness recognition result; wherein the first binary classifier is used to identify whether the to-be-tested object has a defect, and the second binary classifier is used to identify whether the thickness value of the to-be-tested object is within the preset thickness threshold range.

3. The object detection method of claim 1, wherein When the defect recognition result is used to represent that the to-be-tested object has a surface defect, inputting the defect recognition result into a second convolutional neural network model to obtain defect information of the to-be-tested object, wherein the second convolutional neural network model is used to identify the defect information of the to-be-tested object.

4. The object detection method according to claim 3, wherein The inputting the defect recognition result into the second convolutional neural network model to obtain the defect information of the to-be-tested object comprises: extracting defect feature data in the defect recognition result; extracting defect position information in the defect feature data according to a region prediction network in the second convolutional neural network model, performing full convolution processing on the defect position information, classifying pooling features of the defect position information after the full convolution processing to obtain a weight, performing convolution processing on the weight and a preset feature quantity to obtain defect label semantics; fusing the defect position information and the defect label semantics, and performing convolution on the fused defect position information and defect label semantics to obtain label semantics of the to-be-tested object; predicting the label semantics of the to-be-tested object to obtain a predicted defect category; determining the defect position information and the predicted defect category as the defect information.

5. The object detection method according to any one of claims 1 to 4, characterized in that, The extracting the first feature signal in the ultrasonic signal and the second feature signal in the pulse eddy current signal comprises: extracting the first feature signal in the ultrasonic signal and the second feature signal in the pulse eddy current signal through a self-encoder, wherein the self-encoder is a self-encoder with a hidden layer with a preset threshold number of layers and a sparse penalty function.

6. An object detection device, characterized by, The method comprises a controller, a first electromagnetic ultrasonic magnet, a first electromagnetic ultrasonic coil, a second electromagnetic ultrasonic magnet, a second electromagnetic ultrasonic coil, a third electromagnetic ultrasonic coil, a first pulse eddy current coil, a second pulse eddy current coil and a ferrite shielding layer. The controller is connected with the first electromagnetic ultrasonic coil, the second electromagnetic ultrasonic coil, the third electromagnetic ultrasonic coil, the first pulse eddy current coil and the second pulse eddy current coil respectively, the first pulse eddy current coil is arranged between the first electromagnetic ultrasonic magnet and the third electromagnetic ultrasonic coil, the second pulse eddy current coil is arranged between the third electromagnetic ultrasonic coil and the second electromagnetic ultrasonic magnet, the first electromagnetic ultrasonic coil is arranged below the first electromagnetic ultrasonic magnet, the second electromagnetic ultrasonic coil is arranged below the second electromagnetic ultrasonic magnet, and the third electromagnetic ultrasonic coil is arranged in the ferrite shielding layer. The controller is configured to control the first electromagnetic ultrasonic coil and the second electromagnetic ultrasonic coil to emit detection signals to a to-be-detected object, receive ultrasonic signals transmitted by the third electromagnetic ultrasonic coil, control the first pulse eddy current coil to emit detection signals to the to-be-detected object, and receive pulse eddy current signals input by the second pulse eddy current coil, or control the second pulse eddy current coil to emit detection signals to the to-be-detected object, and receive pulse eddy current signals input by the first pulse eddy current coil, so as to realize the object detection method according to any one of claims 1-5.

7. An object detection apparatus characterized by comprising: The method comprises the following steps of: The first determining module is configured to determine ultrasonic signals and pulse eddy current signals of the to-be-detected object. The signal extracting module is configured to extract first characteristic signals in the ultrasonic signals and second characteristic signals in the pulse eddy current signals. The identifying module is configured to input the first characteristic signals and the second characteristic signals into a first convolutional neural network model to obtain a defect identification result and a thickness identification result of the to-be-detected object, wherein the first convolutional neural network model is configured to identify whether the to-be-detected object has a defect and whether the thickness value is within a preset thickness threshold range.

8. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method according to any one of claims 1-5.

9. A controller characterized by comprising: The computer program is executed by the processor to implement the steps of the method according to any one of claims 1-5. The computer program is executed by the processor to implement the steps of the method according to any one of claims 1-5. ​ 10. A computer program product comprising a computer program, characterized in that, ​

Images