Method for generating training data set, training data set and computing unit, computer program product

Abstract

The invention relates to a computer-implemented method (300) for generating a training data set (143) for training artificial intelligence for operating a measuring device (100), in particular a wall diagnostic device, comprising: receiving (301), by a classifier module (183), sensor data (173) of at least one sensor unit (101) of the measuring device (100); generating (303), by the classifier module (183), a submerged space representation (189) of the sensor data (173) in the submerged space (195); ascertaining (305) a distance (D) between the submerged space representations (189) in the submerged space (195) by a distance ascertaining module (193); classifying (307), by a classification module, the sensor data (173) with respect to a wall type (123) of the wall (105) and / or an object position (115) and / or an object type (117) of the object (113) on the basis of the distance (D) of the submerged space representation (189) of the sensor data (173) in the submerged space (195); and summarizing (309) the classified sensor data (173) into a training data set (143). The invention also relates to a training data set, a computing unit and a computer program product.

Description

Technical Field

[0001] The present invention relates to a method for generating a training data set for training artificial intelligence for a wall diagnosis device.

[0002] The invention further relates to a training data set, a computing unit and a computer program product. Background Art

[0003] Diagnostic devices for diagnosing walls and for detecting objects formed in walls are known from the prior art. Summary of the Invention

[0004] The object of the present invention is to provide an improved method for generating a training data set for training an artificial intelligence of a wall diagnosis device.

[0005] This object is achieved by the method according to the invention. Advantageous embodiments are the subject of further developments.

[0006] According to one aspect, a computer-implemented method for generating a training data set for training an artificial intelligence for operating a measuring device, in particular a wall diagnostic device, is provided, wherein the method comprises:

[0007] receiving sensor data from at least one sensor unit of the measuring device by a classifier module, wherein the sensor data map a wall to be diagnosed having a wall type, wherein objects belonging to the object type are arranged at object positions in the wall;

[0008] The classifier module generates a latent space representation of the sensor data in the latent space Therein, the latent space representation of the sensor data is constructed as a dimensionality-reduced representation of the sensor data;

[0009] The distance calculation module calculates the distance between the latent space representations in the latent space;

[0010] Classifying the sensor data by a classification module with respect to a wall type of a wall and / or an object position and / or an object type of an object based on a distance between latent space representations of the sensor data in the latent space, wherein the sensor data are classified as belonging to a common class if the distance between the individual latent space representations in the latent space is less than or equal to a predefined limit value; and

[0011] Aggregate the classified sensor data into a training dataset.

[0012] This provides the following technical advantages: an improved method for generating a training dataset for training artificial intelligence for operating a measuring device, particularly a wall diagnostic device, can be provided. To this end, a classifier module first receives sensor data from at least one sensor unit of the measuring device. The sensor data represents a wall to be diagnosed, which has a wall type, and objects that may be located within the wall. Based on the sensor data, the classifier module generates a latent space representation of the sensor data. The latent space representation is a dimensionality-reduced representation of the sensor data. Therefore, the latent space representation encodes the sensor data and can be presented in vector form.

[0013] Furthermore, a distance determination module determines distances in the latent space between the latent space representations. Based on the determined distances between the latent space representations within the latent space, a classification module classifies the sensor data with respect to the wall type of the wall to be diagnosed and / or the object position and / or object type of an object arranged in the wall.

[0014] Here, classification is performed such that a latent space representation in the latent space is assigned the first classification if its distance from a group of other latent space representations that have already been assigned the first classification is less than or equal to a predetermined distance. Thus, determining the distance of the latent space representation in the latent space allows for a more precise classification of the sensor data represented by the latent space representation with respect to the wall type of the represented wall or the object position and / or object type of an object.

[0015] Therefore, determining the distances in the latent space representation allows for simple and reliable automatic classification of sensor data with reference to classification features. The correspondingly classified sensor data is then aggregated into a training dataset. This avoids the need for tedious manual classification of the reclassified sensor data.

[0016] Classified sensor data, within the meaning of the present invention, refers to sensor data for which the corresponding classification is known with respect to the classification characteristics. Classification characteristics may, for example, be the wall type of a wall and / or the object position and / or object type of an object. Thus, for classified sensor data, the wall type of the wall mapped by the sensor data and / or the object position and / or object type of the object mapped by the sensor data are known. Classification, in this context, involves identifying sensor data with respect to the classification characteristics. Sensor data is identified via classification as to the wall type of the wall mapped by the sensor data or the object position and / or object type of the object mapped by the sensor data.

[0017] Here, the unclassified sensor data may be a plurality of measurements from a plurality of measuring devices, wherein the measurements are performed on a plurality of different walls with different objects and different wall types. These measurements may be primarily for generating sensor data for a training dataset, or may be performed by a user during use of a measuring device.

[0018] According to one embodiment, the method further comprises:

[0019] If the distance of the corresponding latent space representation to the latent space representations of other sensor data of the training data set in the latent space is greater than or equal to a predefined second limit value, these sensor data are removed from the training data set or are not considered for the training data set.

[0020] This provides the following technical advantage: sensor data that cannot be assigned to any object class based on distance determination can be excluded from the training dataset based on the distance representation of the latent space, or can be disregarded for the training dataset. If the distance between a latent space representation and an already classified latent space representation in the latent space is greater than or equal to a predefined second limit value, the sensor data represented by the corresponding latent space cannot be assigned to any class of already classified sensor data. Such unclassifiable sensor data may, for example, be based on erroneous measurements and be erroneous sensor data. By excluding such sensor data from the training dataset or removing it from the training dataset, the quality of the resulting training dataset can be improved.

[0021] According to one embodiment, the method further comprises:

[0022] Generating, by the classifier module, generated sensor data for a training dataset, wherein the generating comprises:

[0023] generating, by the classifier module, a latent space representation having a predefined distance to a preselected latent space representation in the latent space, wherein the preselected latent space representation is sensor data of a preselected classification; and

[0024] generating, by a classifier module, sensor data represented by the generated latent space representation, wherein the generated sensor data has the preselected classification; and

[0025] Add the generated sensor data to the training dataset.

[0026] This results in the following technical advantage: the sensor data for the training data set can be generated by the classifier module, thereby avoiding complex measurements for recording the sensor data for the training data set.

[0027] Instead, corresponding sensor data can be generated simply by implementing a classifier module, which is not based on actual measurements by the sensor unit of the measuring device. To this end, the classifier module first generates a latent space representation that is at a predefined distance from a preselected latent space representation in the latent space. As mentioned above, the latent space representation is a vector representation that is reduced in dimensionality compared to the actual sensor data and encodes information about the sensor data.

[0028] By generating a corresponding vector representation or latent space representation in the latent space, corresponding sensor data can be represented, which was not actually recorded by the measurement. By generating a latent space representation at a predefined distance from a preselected latent space representation, it is possible to ensure that the newly generated latent space representation has the preselected classification of the preselected latent space representation. This ensures that the sensor data represented by the generated latent space representation also has the preselected classification.

[0029] Based on the correspondingly generated latent space representation, sensor data correspondingly represented by the latent space representation can then be generated by decoding the information of the latent space representation into the form of sensor data. This allows the generation of classified sensor data that is not based on measurements actually performed by the sensor unit and has a predefined classification. Thus, sensor data can be generated in a customized manner according to the desired classification. Consequently, the training dataset can be expanded with any amount of newly generated sensor data and thus scaled to any size, without requiring complex measurements to generate correspondingly convincing sensor data.

[0030] Furthermore, the training dataset can be extended with sensor data that have the desired classification, i.e., sensor data maps that are not well represented in the existing training dataset. Thus, the sensor data can be generated precisely and customized to the requirements of the training dataset, so as to generate a training dataset that is as comprehensive and balanced as possible, in which as many different wall categories, object locations, and object types as possible are represented.

[0031] According to one embodiment, the classification further comprises:

[0032] The sensor data of the training dataset are labeled corresponding to the classification.

[0033] This provides the following technical advantage: sensor data can be generated by classifying and simultaneously labeling sensor data. Labeled sensor data corresponds to annotated sensor data in which each class is labeled with a corresponding label. This further improves the quality of the resulting training dataset.

[0034] According to one embodiment, the method further comprises:

[0035] The sensor data of the training dataset are visualized by visualizing the corresponding latent space representation of each sensor data by a visualization module.

[0036] This results in the following technical advantages: Based on the visualized latent space representations, due to their respective arrangement and spacing relative to one another, objects represented by these latent space representations can be assigned to common object classes and / or object domains. Object classes are definitions of different object types, and object domains are collections of multiple object classes.

[0037] According to one embodiment, the distance is defined as the Euclidean distance.

[0038] This results in the technical advantage that a simple distance determination is possible.

[0039] According to one specific embodiment, the classifier module is designed as a correspondingly trained artificial intelligence which is configured to generate a corresponding latent space representation based on the sensor data and / or to generate corresponding sensor data based on the latent space representation.

[0040] This provides the following technical advantages: a powerful and reliable classifier module with the above characteristics can be provided.

[0041] According to one embodiment, the classifier module is designed as a generative adversarial network (GAN), in particular as an autoencoder having an encoder module and a decoder module.

[0042] This provides the following technical advantages: a powerful and reliable classifier module can be provided.

[0043] According to one embodiment, the received sensor data includes tagged and / or untagged sensor data.

[0044] This results in the technical advantage that the method can be applied both to already labeled sensor data and to sensor data which have not been labeled to date.

[0045] According to one embodiment, the sensor data includes radar data of a radar sensor and / or data of an inductive sensor and / or an eddy current sensor and / or a capacitive sensor and / or an AC current sensor and / or a nuclear magnetic resonance (NMR) sensor and / or an ultrasonic sensor.

[0046] This provides the technical advantage that, in addition to the information from the radar data of the radar sensor unit, additional information can be incorporated into the wall diagnosis through further sensor data from additional sensors, each configured to detect a different physical measured variable. This additional information, which is preferably complementary to the information from the radar data of the radar sensor unit, further increases the accuracy of the wall diagnosis or object detection.

[0047] According to one specific embodiment, the sensor data are also classified with reference to the object depth and / or the object extent of the object.

[0048] This results in the technical advantage that additional information from the training data set relating to further features of the object arranged in the wall to be inspected can be taken into account in the training of the diagnosis module and thus in its wall diagnosis.

[0049] According to one embodiment, the object category of the object type of the object includes: metal / non-metallic objects, cables for low voltage, cables with single-phase AC signals, cables with multi-phase AC signals, wooden brackets, metal brackets, plastic pipes, plastic pipes filled with water (such as tap water pipes), plastic pipes not filled with water (such as sewage pipes), and / or the wall type category of the wall type of the wall includes: concrete wall, light structure / dry structure wall, brick wall and / or brick stone of the wall, floor heating facility, wall heating facility.

[0050] This results in the following technical advantages: a large number of different objects of different object types can be identified or classified. The measuring device or diagnostic module can be trained to detect and classify typical objects built into building walls. This enables particularly precise wall diagnostics, in which detected objects can be precisely and unambiguously assigned to the corresponding object class.

[0051] Precise object classification and providing the user with corresponding classification information via the display unit enables the most convincing possible wall diagnosis. Knowing not only where the object is located within the wall but also the object type of the detected object, the user can accordingly decide how to further process the wall with respect to the detected object. Therefore, providing the object type of the detected object classification is a key area of ​​wall diagnosis, as the user can adjust the planned wall processing accordingly based on the given object type.

[0052] According to one aspect, a training data set for training artificial intelligence of a measuring device for wall diagnostic technology is provided, wherein the training data set is generated by the method for generating a training data set according to one of the above embodiments, and the training data set is used to train artificial intelligence for operating the measuring device.

[0053] According to one aspect, a computing unit is provided which is configured to implement a method for generating a training data set for training an artificial intelligence for operating a measuring device and / or a method for training an artificial intelligence of a measuring device according to one of the above embodiments.

[0054] According to one aspect, a computer program product is provided comprising instructions which, when executed by a data processing unit, cause the data processing unit to implement a method for generating a training data set and / or a method for training an artificial intelligence of a measuring device according to one of the above-described embodiments, the training data set being used to train an artificial intelligence for operating a measuring device. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Embodiments of the present invention will be described with reference to the following drawings. The drawings show:

[0056] Figure 1 A schematic diagram of a measuring device according to one embodiment;

[0057] Figure 2 Another schematic diagram of a measuring device according to another embodiment;

[0058] Figure 3 Another schematic diagram of a measuring device according to another embodiment;

[0059] Figure 4 Schematic diagram of a measurement of a measuring device according to one embodiment;

[0060] Figure 5 Another schematic diagram of a measuring device according to another embodiment;

[0061] Figure 6A schematic diagram of a system for generating a training dataset according to one embodiment;

[0062] Figure 7 A flowchart of a method for generating a training data set according to one embodiment;

[0063] Figure 8 Another flow chart of a method for generating a training data set according to another embodiment;

[0064] Figure 9 Another flow chart of a method for generating a training data set according to another embodiment;

[0065] Figure 10 Another flow chart of a method for generating a training data set according to another embodiment;

[0066] Figure 11 Another flow chart of a method for generating a training data set according to another embodiment;

[0067] Figure 12 Another flow chart of a method for generating a training dataset according to another embodiment; and

[0068] Figure 13 Schematic diagram of a computer program product. DETAILED DESCRIPTION

[0069] Figure 1 A schematic diagram of a measuring device 100 according to one embodiment is shown.

[0070] The present invention relates to a measuring device, and in particular to a wall diagnostic device for inspecting a wall 105 to be processed. Wall diagnostic devices for detecting objects arranged in a wall are known in the prior art. Such devices allow a user to inspect the wall to be processed based on the objects arranged in the wall, so that planned work, such as drilling a hole in the wall, can be carried out based on this information, in a manner that avoids damaging the objects arranged in the wall.

[0071] In the embodiment shown, the measuring device 100 comprises a housing 150 having a handle 152 for a user to hold the measuring device 100 , a display unit 111 for displaying a diagnosis result 109 of a wall diagnosis, and operating elements 154 for switching the measuring device 100 to different operating modes.

[0072] According to the present invention, measuring device 100 comprises at least one radar sensor unit 101. Radar sensor unit 101 can transmit radar signals in the direction of a wall 105 to be inspected and can receive radar signals reflected by wall 105.

[0073] Radar sensor unit 101 may be designed, for example, as a narrowband radar detector device in the frequency range of 2.4 GHz to 2.4835 GHz or as an ultra-wideband radar detector device in the frequency range of 1.8 GHz to 5.8 GHz.

[0074] In addition, to perform wall diagnostics, measuring device 100 includes a diagnostic module 107, which may be implemented on computing unit 151 of measuring device 100. Diagnostic module 107 is configured to perform appropriate diagnostics of the wall to be inspected based on radar data 103 from radar sensor unit 101. Radar data 103 from radar sensor unit 101 maps wall 105 to be inspected and objects 113 that may be located within wall 105.

[0075] The wall diagnosis performed by the diagnostic module 107 includes at least object recognition. Object recognition includes detecting and classifying objects 113 located in the wall 105. Object detection includes at least determining an object position 115. The object position describes the position of the object located in the wall 105 relative to the reference system determined by the measuring device 100. Classifying the detected object 113 includes at least determining an object type 117 of the detected object 113.

[0076] The diagnostic results of the wall diagnosis thus determined, i.e., at least the determined object position 115 and / or the determined object type 117 of the object 113 arranged in the wall 105, are then displayed to the user of the measuring device 100 on the display unit 111 of the measuring device 100. The display unit 111 can be designed, for example, as a corresponding display screen, and the diagnostic result 109 can be displayed visually. In addition, the display of the diagnostic result 109 can be supported by acoustic and / or haptic signals. The haptic signal can be implemented, for example, by a corresponding vibration signal.

[0077] Here, object 113 can be displayed, for example, by a corresponding symbol on a display screen. Object 113 can be displayed on a display screen at a corresponding object position 115. Object extent 121 can be visualized by the corresponding size of the displayed symbol. Individual object types 117 of object 113 can be visualized using corresponding terms, or by the color background of the symbol, or by a specific shape of the symbol representing object 113.

[0078] Alternatively, the wall diagnosis can also additionally include the determination of a wall type 123 in the form of a wall type classification of the wall 105 to be inspected. Wall type 123 describes the individual types of wall 105 to be inspected. For example, the wall type can be assigned to a corresponding wall type category, which can include: concrete wall, lightweight / dry-built wall, brick wall and / or brick and stone wall, floor heating system, wall heating system, or similar wall types found in buildings.

[0079] According to one embodiment, diagnostic module 107 is further configured to determine an object depth 119 of object 113 within wall 105 based on radar data 103. Object depth 119 is defined as the distance of an object formed within wall 105 from the surface of wall 105. On the object side, this distance can be defined, for example, with reference to the object surface or to the object's center point. The distance to the surface of wall 105 describes the shortest distance defined by a direction perpendicular to the surface of wall 105.

[0080] According to one embodiment, diagnostic module 107 is further configured to determine an object extent 121 of object 113 in at least one predefined direction based on radar data 103. Object extent 121 of object 113 describes the spatial extent of object 113 in at least one spatial direction, preferably in two spatial directions, and particularly preferably in three spatial directions. Thus, object 113 can be described as a one-dimensional, two-dimensional, or three-dimensional object 113.

[0081] In typical use, measuring device 100 is placed on the surface of a wall 105 to be inspected. Radar signals are transmitted in the direction of wall 105 via radar sensor unit 101, and radar signals reflected by wall 105 or objects 113 located behind the wall are received. Diagnostic module 107 performs the aforementioned wall diagnosis on radar data 103 from radar sensor unit 101 and determines corresponding diagnostic results 109.

[0082] Diagnosis result 109 may include, for example, object position 115 and / or object type 117 of object 113 arranged in wall 105. Alternatively or additionally, diagnosis result 109 may also include wall type 123 of wall 105 and / or object depth 119 and / or object extension 121 of object 113.

[0083] The thus configured diagnosis result 109 can then be displayed to the user of the measuring device 100 on the display unit 111 of the measuring device 100. The display unit 111 can be designed as a corresponding display screen, for example. The diagnosis result 109 can be displayed on the display unit 111 in graphical form or in text form.

[0084] According to one embodiment, measuring device 100 further includes a motion detection unit 141. Motion detection unit 141 can be used to detect the movement of measuring device 100 relative to wall 105. To this end, motion detection unit 141 can include, for example, at least one roller element. When the roller element is placed on the surface of wall 105, the movement of measuring device 100 relative to wall 105 can be detected when measuring device 100 rolls along motion direction 153 due to the rolling of the roller element. Alternatively, motion detection unit 141 can have another configuration that can detect the relative movement of measuring device 100 relative to wall 105.

[0085] By moving measuring device 100 relative to wall 105, radar data 103 of radar sensor unit 101 can be recorded for a plurality of different positionings of measuring device 100 relative to wall 105. This allows for the examination of wall 105 over a larger spatial area than that given by the range of action of radar sensor unit 101. This allows for the detection of objects 113 having a larger spatial extent than the range of action of radar sensor unit 101.

[0086] While measuring device 100 is moving in direction of movement 153, sensor data 103 from radar sensor unit 101 can be continuously recorded. Based on these radar data 103, wall diagnosis can be evaluated by diagnostic module 107 while measuring device 100 is moving in direction of movement 153. This enables accelerated wall diagnosis that takes into account the positioning of measuring device 100 relative to wall 105.

[0087] Depending on its embodiment, diagnostic module 107 is designed as a correspondingly trained artificial intelligence 125. Artificial intelligence 125 is trained to perform the aforementioned wall diagnosis based on radar data 103 from radar sensor unit 101 and to determine at least object position 115 and object type 117 of object 113 located in wall 105. Object classification or determination of object type 117 here includes assigning detected object 113 to a predefined object class.

[0088] Here, object categories may include: metal / non-metal objects, cables for low voltage, cables with single-phase AC signals, cables with multi-phase AC signals, wooden supports, metal supports, plastic pipes, plastic pipes filled with water (such as tap water pipes), plastic pipes not filled with water (such as sewage pipes), or other elements that are typically built into the walls of buildings.

[0089] Furthermore, artificial intelligence 125 is trained to determine wall type 123 of wall 105 to be inspected based at least on radar data 103 from radar sensor unit 101. Possible wall types 123 may include: concrete walls, lightweight / dry-built walls, brick walls and / or individual brickwork, floor heating systems, wall heating systems, or other common wall types constructed in buildings.

[0090] According to one embodiment, in addition to radar sensor unit 101, measuring device 100 may also include additional sensors, by means of which additional physical variables can be detected. For example, measuring device 100 may include an inductive sensor, an eddy current sensor, a capacitive sensor, an AC current sensor, a nuclear magnetic resonance (NMR) sensor, an ultrasonic sensor, or other sensors typically incorporated into wall diagnostic equipment.

[0091] Diagnostic module 107, in particular a correspondingly trained artificial intelligence 125, may be configured to perform the aforementioned wall diagnosis based on radar data 103 from radar sensor unit 101 and taking into account additional sensor information from further sensors. For this purpose, the additional information from the further sensors may be used, in particular, for object recognition of objects 113 located in wall 105. The additional sensor information may enable improved detection and, if necessary, improved classification of objects 113.

[0092] The material of object 113 , for example metallic or non-metallic, can be refined and classified, for example, by using additional sensor information.

[0093] Figure 2 A further schematic diagram of a measuring device 100 according to another specific embodiment is shown.

[0094] In the illustrated embodiment, in addition to the diagnostic module 107, the measuring device 100 also includes a preprocessing module 127. For wall diagnosis, the measuring device 100 first receives radar data 103 from the radar sensor 101. The received radar data 103 is preprocessed by the preprocessing module 127. Through the preprocessing by the preprocessing module 127, the radar data can be converted into a corresponding data structure required by the diagnostic module 107 for wall diagnosis, for example.

[0095] As described above, the diagnostic module 107 generates the diagnostic result 109 during the wall diagnosis. Here, the diagnostic result 109 may include, for example, the object position 115 and / or the object type 117 and / or the object depth 119 and / or the object extension 121 of the object 113 located in the wall 105 to be inspected, and / or the wall type 123 of the wall 105 to be inspected. The generated diagnostic result 109 may then be displayed on the display unit 111 of the measuring device 100.

[0096] According to one embodiment, in addition to radar data 103 from radar sensor unit 101, the aforementioned additional sensor information from the additional sensors is also considered in the wall diagnosis of diagnosis module 107. Preprocessing module 127 can accordingly preprocess the additional sensor information.

[0097] In the illustrated embodiment, the diagnosis module 107 includes a wall type classification module 129 and an object recognition module 131. The preprocessing module 127 includes a first preprocessing module 135 and a second preprocessing module 137. The first preprocessing module 135 includes an S-matrix reduction module 155. The second preprocessing module 137 includes a background correction module 157, an inverse fast Fourier transform 159, and a focus and shift module 161. When the radar data 103 is preprocessed by the preprocessing module 127, the radar data 103 is first preprocessed by the first preprocessing module 135 and the S-matrix reduction module 155 included in the first preprocessing module.

[0098] Here, the first preprocessing module 135 generates input data 133 based on the sensor data 103. This input data 133 serves as input data for the wall type classification module 129. Here, the wall type classification module 129 classifies the wall 105 to be inspected into a wall type based on the input data 133 and generates wall type information 139. The wall type information 139 includes the wall type 123 of the wall 105 to be inspected, which is determined in the wall type classification.

[0099] Subsequently, second preprocessing module 137 performs preprocessing based on radar data 103 and wall type information 139. Background correction 157 of radar data 103 is performed here, taking into account wall type 123 ascertained in wall type information 139. Depending on wall type 123 of wall 105 to be inspected, different effects may occur on radar data 103.

[0100] Background correction 157 can correct for these effects, which are primarily based on the respective wall type 123 and can affect object recognition. After background correction, further preprocessing can be performed by performing an inverse fast Fourier transform 159 or focusing and shifting 161, and new input data 133 for object recognition module 131 can be generated. Based on the input data 133 provided by second preprocessing module 137, object recognition module 133 recognizes objects 113 located in wall 105 to be inspected and determines at least object position 115 and object type 117 of the respective object 113. Object recognition module 131 can also determine object depth 119 and object extent 121.

[0101] According to one specific embodiment, the diagnostic module is further configured to determine an object depth of an object within the wall based on the radar data, wherein the object depth is defined by a distance of the object constructed in the wall from a surface of the wall.

[0102] Preprocessing is optional. Depending on the algorithm used by diagnostic module 107, completely unprocessed radar echoes with different frequencies can be used as radar data 103 and as input data for diagnostic module 107. Alternatively, radar data 103 processed through multiple steps can be preprocessed. Preprocessing steps include, for example, signal transformation from frequency space to time or range space, background removal, noise reduction, and normalization. For radar data 103 in complex form, only absolute values ​​can be processed. Alternatively or additionally, phase information can be considered.

[0103] Figure 3 A further schematic diagram of a measuring device 100 according to another specific embodiment is shown.

[0104] In the embodiment shown, the diagnostic module 107 comprises a plurality of processing paths 102 running in parallel. In each processing path 102 there is: a pre-processing module 127; a diagnostic module 107, which comprises, for example, a Figure 2 The wall type classification module 129 and / or object recognition module 131 described in the embodiment; and the post-processing module 163.

[0105] exist Figure 3In the figure, radar data 103 is primarily shown as input data for wall diagnosis. However, in addition to the radar data shown, additional information from additional sensors can also be used as input data for wall diagnosis. Different information from different sensor types can be processed in separate, parallel processing paths 102, and corresponding wall diagnoses can be performed separately based on the different sensor information. After the wall diagnosis is completed, the summary of the individual analysis results can be combined into a wall diagnosis result 109 via a summary module.

[0106] Alternatively or additionally, different partial aspects of the wall diagnosis may also be performed by different processing paths 102 based on the same sensor information.

[0107] In this case, the various processing paths 102 can process, for example, different radar data 103 that were recorded for different positions of the measuring device 100 relative to the wall 105 during the movement of the measuring device 100 relative to the wall 105. Thus, the radar data 103 that map different areas of the wall 105 and were recorded sequentially in time during the movement of the measuring device 100 relative to the wall 105 can then be processed in the different processing paths 102 by the modules shown.

[0108] In this case, different processing paths perform independent wall diagnostics, which include at least determining an object position 115 and / or an object type 117 of an object 113 arranged in the wall 105 .

[0109] The partial results of the independent wall diagnosis of different areas of the wall 105 provided in each processing path 102 can be aggregated into a coherent Diagnostic result 109. A coherent diagnostic result describes a coherent spatial region that was scanned during the movement of measuring device 100 relative to wall 105 and mapped by the correspondingly recorded radar data 103. Therefore, parallel processing of radar data 103 or additional sensor information 104 from additional sensor elements in different processing paths 102 enables accelerated wall diagnosis.

[0110] Alternatively, different functions of the wall diagnosis can be implemented in different processing paths 102. Thus, for example, wall type classification and determination of wall type 123 of the wall to be inspected 105 can be implemented in one processing path 102. Object recognition of objects 113 arranged in the wall can be implemented in another processing path 102. In this case, object detection with determination of object position 115 and object classification with determination of object type 113 can be implemented in one processing path 102.

[0111] Alternatively, object detection and object classification can also be performed in two separate processing paths 102. In the other processing path 102, object depth determination, i.e., determination of object depth 119 and / or determination of object extent 121, can each be effected. In the aggregation module 165, the various partial results of the wall diagnosis can be aggregated into the corresponding diagnostic result 109.

[0112] Here, the diagnostic module 107 can be divided into different artificial intelligences 125, as has already been described. Figure 2 As shown in the embodiment of FIG. Here, the diagnosis module 107 may include, for example, a wall type classification module 129 and an object recognition module 131. The object recognition module may be further divided into an object detection module and an object classification module. Furthermore, the diagnosis module 107 may also include an object depth determination module and an object extension scale module, which are configured to determine an object depth 119 and an object extension scale 121, respectively.

[0113] The corresponding modules can each be designed as a separate artificial intelligence 125, such as a neural network. Alternatively, the different modules can form parts of an overall artificial neural network, which are connected to form an overall neural network according to structures known from the prior art.

[0114] Figure 4 A schematic diagram shows a measurement of the measuring device 100 according to one embodiment.

[0115] For preprocessing, radar data 103 or additional sensor information 104 from other sensors is normalized, particularly for numerical stabilization in subsequent steps performed by diagnostic module 107 during wall diagnosis. For example, amplitude and / or offset compensation can be performed for this purpose. Furthermore, radar data 103 can be filtered to reduce interfering elements, and the corresponding sensor data can be downsampled to reduce the data rate. Furthermore, radar data 103 or additional sensor information 104 can be transformed into the required frequency or time range. Methods known from the prior art can be used for this purpose.

[0116] Furthermore, recorded radar data 103 or additional sensor information 104 can be divided into time or space windows 167. Here, time windows 167 can be generated by recording radar data 103 or additional sensor information or preprocessed radar data 103 over fixed time intervals. Conversely, space windows 167 can be generated by assigning radar data 103 or additional sensor information 104 to a position of measuring device 100 relative to wall 105 in direction of movement 153.

[0117] Figure 4FIG. a) shows such a data matrix resulting from the above steps. The data matrix shown in FIG. a) in window 167 shows a plurality of sensor data, which may include, for example, radar data 103 or additional sensor information 104 from other sensors, and is plotted along a frequency channel axis 171 and along a space / time axis 169.

[0118] Here, the width of the time window 167 can be selected so that different sampling rates of the sensors can be balanced and a new window 167 can be provided frequently enough so that the diagnostic results 109 of the wall diagnosis can be displayed on the display unit 111 without excessive time delay while the measuring device 100 is performing the measurement or shortly after the measurement by the measuring device 100 ends.

[0119] For this purpose, a data recording rate of 2 to 20 windows per second of sensor data may be advantageous. For the spatial windows, the spatial sampling rate may be selected so as to achieve the desired local accuracy. Here, a sampling rate of 1 mm to 1 cm may be advantageous. This means that for every 1 mm to 1 cm of movement of the measuring device 100 in the direction of motion 153, corresponding sensor data is recorded.

[0120] The width of spatial window 167 can be selected such that coherent information about object 113 is contained in one window. Widths of 1 cm to 20 cm for each spatial window 167 may be advantageous. This results in 4 to 100 measured values ​​per window 167. This enables efficient further algorithmic processing of the corresponding recorded radar data 103 or additional sensor information by diagnostic module 107.

[0121] Here, once one or more sampling points are available, another time window 167 or space window 167 may be provided.

[0122] The diagnostic module 107 can be adapted so that as input data, for example also Figure 3 The input data of each processing path 102 in the embodiment of the present invention is recorded as a matrix corresponding to the window size of each space or time window 167. Here, according to Figure 2 In an embodiment, the corresponding input data may include respectively pre-processed sensor data, namely radar data 103 and additional sensor information 104 of the additional sensor.

[0123] As described above, the wall diagnosis can be performed by the diagnostic module 107 based on a correspondingly trained artificial intelligence. Alternatively, different processing paths can also be calculated by a rule-based algorithm. Artificial intelligence and rule-based algorithms can also be combined within the processing path 102 in a parallel or series connection.

[0124] The wall diagnosis result 109 can be represented as a numerical value, a vector, or a matrix. Furthermore, the probability of detection can be given for object detection, or the probability of a given object or wall type classification can be given for wall type classification or object classification. The same approach applies to position and / or depth determination, for which corresponding probability values ​​can also be given.

[0125] If, in addition to radar data 103 , additional sensor information from other sensor types is processed in a processing path 102 , these radar data and additional sensor information can be either aggregated within artificial intelligence 125 or combined by rule-based combination.

[0126] In the post-processing of each processing path 102, i.e. Figure 3 In the post-processing of the embodiment in , the summing module 165 can summarize the multiple algorithm results based on the multiple windows 167. This summing can be achieved in particular by taking a multiplier, summing, or by multiplying consecutive probability values.

[0127] Furthermore, by clustering a plurality of results, for example, a plurality of objects detected close to one another, it is possible to identify which objects are the same object, so that these objects are not mistakenly identified multiple times.

[0128] A weighting function 177 can also be multiplicatively applied when combining the results from multiple windows 167. Advantageously, the partial diagnostic results 175, which correspond to the corresponding data points in space, can be weighted with reference to their position relative to the center point of the respective window 167. This is illustrated by way of example in FIG. b), where individual partial diagnostic results 175 are weighted with reference to the center point of the illustrated window 167 according to the illustrated weighting function 177.

[0129] According to one embodiment, the result of one processing path 102 after post-processing 163 may influence the expansion of another processing path 102s. Here, a weighting parameter may be adjusted, which may be related to the corresponding result from the processing path 102 for each window.

[0130] For example, the result of the object classification for defining the object type 117 of an object arranged in the wall 105 can be used to increase the weight of the wall type classification for determining the wall type 123 of the corresponding wall 105 at locations where there are no objects 113 in post-processing, because the corresponding radar data 103 at these locations are less affected by the reflection of the object 113.

[0131] Figure 5 A further schematic diagram of a measuring device 100 according to another specific embodiment is shown.

[0132] Figure 5 Graphs a) and b) of FIG. 1 show two different alternatives for the joint data processing of radar data 103 and additional sensor information 104 by diagnostic module 107 .

[0133] FIG. a) illustrates the joint processing of radar data 103 and additional sensor information 104 from an additional sensor by diagnostic module 107. To this end, radar data 103 and additional sensor information 104 are jointly used as input data for diagnostic module 107, which is designed as an artificial intelligence, particularly an artificial neural network. Diagnostic module 107 includes a plurality of convolutional layers 108 and a plurality of dense layers 106. Radar data 103 and additional sensor information 104 are jointly processed as input data via these convolutional layers 108 and dense layers 106. Based on this, the aforementioned diagnostic result 109 is generated as output data from diagnostic module 107.

[0134] In contrast, in Figure b), radar data 103 and additional sensor information 104 are used as separate input data for diagnosis module 107. Diagnosis module 107 becomes a plurality of processing paths 102. Processing paths 102 each include a plurality of convolutional layers 108 and at least one dense layer 106. In different processing paths 102, diagnosis module 107 separately creates a wall diagnosis based on radar data 103 or additional sensor information 104.

[0135] The partial results of the partial diagnosis of the different processing paths 102 are combined in an additional connection layer 148 and then fed to the final dense layer 106. The output data of the diagnosis module 107 correspond to the diagnosis result 109 described above.

[0136] A correspondingly designed diagnostic module 107 is configured to carry out the above-described wall diagnosis with the above-described features based on the radar data 103 and the additional sensor information 104 .

[0137] In the embodiment shown, the diagnosis module 107 is designed as an artificial neural network, in particular as a convolutional network. A corresponding network structure with convolutional layers 108, dense layers 106 and connection layers 148 is known in the prior art.

[0138] Figure 6 A schematic diagram of a system 600 for generating a training dataset 143 according to one embodiment is shown.

[0139] exist Figure 6 FIG. a) of FIG. 1 shows a system 600 for generating a training data set 143. System 600 includes at least one classifier module 183. Classifier module 183 is configured to generate training data set 143 based on sensor data 173 from at least one sensor unit 101 of a measuring device 100, in particular a wall diagnostic device. Sensor data 173 here maps a wall 105 to be diagnosed and an object 113 arranged in wall 105.

[0140] Sensor data 173 can be based on measurements of sensor units 101 of at least one measuring device 100 or of a plurality of different measuring devices 100. These measurements can be performed on a plurality of different walls 105 having a plurality of different objects 113. These measurements can be performed, for example, specifically for the purpose of generating training data set 143. Alternatively or additionally, these measurements, or correspondingly recorded sensor data 173, can come from operating measuring devices 100 of a plurality of different users.

[0141] According to the present invention, the classifier module is configured to classify sensor data 173 with respect to at least one classification feature and, based thereon, generate classified sensor data 173. The classification feature may include a wall type 123 of a wall 105 mapped by the sensor data 173 and / or an object position 115 or object type 117 of an object 113 arranged in the mapped wall 105.

[0142] According to the present invention, the classifier module 183 is configured to generate a latent space representation 189 of the latent space 195 based on the received sensor data 173. Here, the latent space representation is a dimensionality-reduced representation of the sensor data 173. The latent space representation 189 is typically constructed as a vector representation.

[0143] For this purpose, the classifier module 183 can be provided as a correspondingly trained artificial intelligence, which is designed to generate a corresponding latent space representation 189 based on the sensor data 173 .

[0144] In the embodiment shown, the classifier module 183 comprises an autoencoder having an encoder module 185 and a decoder module 187. The autoencoder can be constructed and trained according to autoencoder modules known from the prior art.

[0145] In the embodiment shown, the system 600 further comprises a distance determination module 193 . The distance determination module 193 is configured here to calculate distances D between different latent space representations 190 in the latent space 195 .

[0146] According to one embodiment, the distance determination module 193 is part of the classifier module 183 .

[0147] FIG. b ) shows a schematic diagram of the encoder module 185 of the classifier module 183 generating a latent space representation 189 based on the sensor data 173 .

[0148] In the aforementioned figure, latent space representations 190, 192, and 194 of a latent space 195 are grouped into a first latent space representation 190, a second latent space representation 192, and a third latent space representation 194. The first latent space representations 190 are each spaced apart from one another by a smaller distance than a predefined limit value. The second latent space representations 192 are also spaced apart from one another by a smaller distance than a predefined limit value. The same applies to latent space representations 194, which are also spaced apart from one another by a smaller distance than a predefined limit value.

[0149] The first, second and third latent space representations 190, 192, 194 grouped in this way each describe sensor data 173 having three different classifications, wherein the first latent space representation 190 has a first classification, the second latent space representation 192 has a second classification and the third latent space representation 194 has a third classification with respect to a common classification feature or with respect to a plurality of different classification features.

[0150] FIG. b) also shows a latent space representation 189 whose distance D from each of the first latent space representation 190, the second latent space representation 192, and the third latent space representation 194 is greater than a predefined boundary value. In FIG. b), this is demonstrated by the fact that the latent space representation 189 is arranged outside the frame surrounding the group consisting of the first to third latent space representations 190, 192, and 194.

[0151] Therefore, in the example shown, the latent space representation 189 is not part of the three groups of the first to third latent space representations 190 , 192 , 194 and thus does not have a classification with the first to third latent space representations 190 , 192 , 194 .

[0152] If latent space representation 189 is arranged within the group consisting of the first to third latent space representations 190, 192, and 194, then the latent space representation 189 has the corresponding classification of the corresponding group. In this case, the distance D between the latent space representation 189 and the latent space representation 190, 192, and 194 of the corresponding group is less than or equal to the predetermined boundary value.

[0153] FIG. 2 b) shows only one example. In reality, there may be many such groups of latent space representations 190, 192, and 194 of the same classification in the latent space.

[0154] According to the present invention, system 600 is configured via classifier module 183 to classify sensor data 173 based on distance D between latent space representations 189 determined by distance determination module 193. Here, sensor data 173 are classified if the distance between latent space representation 189, which respectively represents the sensor data, and latent space representation 190, 192, 194 of the already classified sensor data 173 is less than or equal to a predefined limit value.

[0155] Here, by Figure 1 The classification module, not shown in FIG, classifies the sensor data 173 in terms of classification features based on the distance D between the latent space representation 189 and the latent space representations 190 , 192 , 194 representing the already classified sensor data.

[0156] Classification here means assigning each sensor data 173 to be classified to a corresponding object class of classified sensor data 173 based on the determined distances D between latent space representations 189, 190, 192, 194. Here, sensor data 173 are classified at least with respect to wall type 123 of wall 105 to be inspected and / or object position 115 and / or object type 117 of object 113 arranged in wall 105.

[0157] Alternatively or additionally, a classification may be effected for this purpose with regard to object depth 119 and / or object extent 121 of object 113 .

[0158] Here, received sensor data 173 may include tagged or untagged sensor data 173. Here, sensor data 173 may include radar data 103 and / or additional sensor information 104 of additional sensors.

[0159] The distance D between the latent space representations 189 , 190 , 192 , 194 determined by the distance determination module 193 can be defined as the Euclidean distance, for example. Alternatively or additionally, the distance can also be defined according to other known distance metrics.

[0160] According to the present invention, system 600 or classifier module 183 is further configured to combine correspondingly classified sensor data 173 into training data set 143 .

[0161] Furthermore, system 600 or classifier module 183 is configured such that if a distance D between latent space representation 189 representing sensor data 173 determined by distance determination module 193 and latent space representations 190, 192, 194 of classified sensor data 173 is greater than a predefined second limit value, the corresponding sensor data 173 is removed from training data set 143 or is not considered for the training data set, i.e., these sensor data 173 are not integrated into training data set 143. This situation is illustrated in FIG. b).

[0162] Therefore, if the latent space representation representing the corresponding sensor data 173 is too far away from the latent space representation 190, 192, 194 representing the classified sensor data 173, then the corresponding sensor data 173 is classified as not suitable for the training dataset 143 and is not considered in the training dataset 143.

[0163] In this way, erroneous sensor data 173 which cannot be classified according to predefined objects via distance determination of the respective latent space representation 189 can be removed from training data set 143 or not considered for the training data set, so that a homogeneous training data set 143 can be generated.

[0164] According to one specific embodiment, classifier module 183 is further configured to generate generated sensor data 191 via decoder module 187. To this end, latent space representation 189 is first generated. As described above, latent space representation 189 is constructed as a vector representation. Here, latent space representation 189 is generated such that, according to the diagram in FIG. b), these latent space representations 189 are assigned to known groups of latent space representations 190, 192, 194 of classified sensor data 173. The vector representations are generated such that a distance D from the vector representations of the first to third latent space representations 190, 192, 194 is less than or equal to a predetermined limit value.

[0165] By executing the correspondingly trained decoder module 187 based on the thus generated latent space representation 189, correspondingly generated sensor data 191 can be generated. Here, the correspondingly generated latent space representation 189 is a representation of the generated sensor data 191. Here, the correspondingly generated sensor data 191 is automatically classified according to the classification feature because the previously generated latent space representation 189 is generated such that, in latent space 195, the distance D between the corresponding generated latent space representation 189 and the latent space representations 190, 192, 194 of the already classified sensor data 173 is less than or equal to a predefined boundary value. The correspondingly generated sensor data 191 is then integrated into the training dataset 143.

[0166] Figure 7 A flow chart of a method 300 for generating a training data set 143 according to one embodiment is shown.

[0167] To generate training data set 143, classifier module 183 first receives sensor data 173 from at least one sensor unit 101 of measuring device 100 in a first method step 301. Sensor data 173 here represent a wall 105 to be diagnosed having a wall type 123 and an object 113 arranged at an object position 115 in wall 105 and belonging to object type 117.

[0168] Here, the sensor data 173 may include tagged and / or untagged sensor data 173 .

[0169] In a further method step 303 , the classifier module 183 generates a latent space representation 189 based on the sensor data 173 .

[0170] In a further method step 305 , distance determination module 193 determines a distance D between latent space representations 189 in latent space 195 . Distance D can be defined as a Euclidean distance.

[0171] Subsequently, in a further method step 307, the sensor data 173 are classified by a classification module based on the distance D of the latent space representation 189. Here, the classification is performed with respect to the wall type 123 of the wall 105 and / or the object position 115 and / or the object type 117 of the object 113. The classification is performed such that if the distance D of the latent space representation 189 of the sensor data 173 to be classified to the latent space representations 190, 192, 194 of the already classified sensor data is less than or equal to a predefined limit value, the latent space representation 189 of the sensor data 173 to be classified is assigned to the common class of the latent space representations 190, 192, 194 of the already classified sensor data 173.

[0172] In a further method step 309 , sensor data 173 classified in this way via the distance determination between latent space representations 189 , 190 , 192 , 194 are combined in training data set 143 .

[0173] Figure 8 A further flow chart of a method 300 for generating a training data set 143 according to another specific embodiment is shown.

[0174] Figure 8 The implementation in Figure 7 and includes all Figure 7 Features described in .

[0175] In the embodiment shown, if the distance D from the latent space representation 189 to the latent space representation 190, 192, 194 of the classified sensor data 173 in the latent space 195 is greater than or equal to a predefined second limit value, then after the classification or after the distance D between the latent space representations 189, 190, 192, 194 is determined, the corresponding sensor data 173 is removed from the training data set 143 in method step 311 or is not considered for the training data set 143.

[0176] In this case, the first and second limit values ​​can be freely selected by the experimenter depending on the quality of sensor data 173 .

[0177] Figure 9 A further flow chart of a method 300 for generating a training data set 143 according to another specific embodiment is shown.

[0178] Figure 9 The embodiment shown in Figure 7 and includes all Figure 7 Features described in .

[0179] In the embodiment shown, in method step 313 , generated sensor data 191 for training data set 143 are generated by classifier module 183 .

[0180] To this end, in method step 315 , a latent space representation 189 is generated by the classifier module 183 in the latent space 195 at a predefined distance D from a preselected latent space representation 190 , 192 , 194 . In this case, the preselected latent space representation 190 , 192 , 194 is the sensor data 173 that has already been classified with a preselected classification.

[0181] In a further method step 317 , corresponding generated sensor data 191 are generated by classifier module 183 based on latent space representation 189 generated in method step 315 .

[0182] In a further method step 319 , the sensor data 191 generated in this way are added to the training data set 143 .

[0183] Figure 10 A further flow chart of a method 300 for generating a training data set 143 according to another specific embodiment is shown.

[0184] Figure 10 The embodiment shown in Figure 7 and includes all Figure 7 Features described in .

[0185] In the embodiment shown, in order to classify the sensor data in method step 307, sensor data 173 are labeled in method step 321. Here, a label is an identifier of the individual sensor data, which indicates the identification of the corresponding classification of sensor data 173.

[0186] Figure 11 A further flow chart of a method 300 for generating a training data set 143 according to another specific embodiment is shown.

[0187] Figure 11 The embodiment shown in Figure 7 and includes all Figure 7 Features described in .

[0188] In the embodiment shown, sensor data 173 of training data set 143 are visualized in method step 323. The visualization of sensor data 173 is thereby effected by visualizing latent space representations 189, 190, 192, 194, which respectively represent sensor data 173. This visualization is performed by a correspondingly configured visualization module.

[0189] Figure 12 A further flow chart of a method 300 for generating a training data set 143 according to another specific embodiment is shown.

[0190] exist Figure 12 The embodiment shown in Figures 7 to 11 The combination of embodiments is based on and includes all Figures 7 to 11 Features described in .

[0191] Figure 13A schematic diagram of a computer program product 500 is shown, which includes instructions which, when executed by a data processing unit, cause the data processing unit to carry out the method 300 for generating a training data set 143 .

[0192] In the embodiment shown, the computer program product 500 is stored on a storage medium 501. In this case, the storage medium 501 may be any storage medium known from the prior art.

Claims

1. A computer-implemented method (300) for generating a training data set (143) for training an artificial intelligence for operating a measuring device (100), in particular a wall diagnostic device, comprising: Sensor data (173) of at least one sensor unit (101) of a measuring device (100) are received (301) by a classifier module (183), wherein the sensor data (173) represent a wall (105) to be diagnosed having a wall type (123), wherein an object (113) of an object type (117) is arranged at an object position (115) in the wall (105); generating (303) a latent space representation (189) of the sensor data (173) in a latent space (195) by the classifier module (183), wherein the latent space representation (189) of the sensor data (173) is constructed as a dimensionality-reduced representation of the sensor data (173); A distance determination module (193) determines (305) a distance (D) between the latent space representations (189) in the latent space (195); The sensor data (173) are classified (307) by a classification module based on the distance (D) between the latent space representations (189) of the sensor data (173) in the latent space (195) with respect to the wall type (123) of the wall (105) and / or the object position (115) and / or the object type (117) of the object (113), wherein the corresponding sensor data (173) are classified as belonging to a common class if the distance between the latent space representations (189) in the latent space (195) is less than or equal to a predefined limit value; and The classified sensor data (173) is aggregated (309) into a training dataset (143).

2. The method (300) of claim 1, wherein: The method further comprises: If the distance (D) from the latent space representation (189) in the latent space (195) to the latent space representation (189) of other sensor data (173) of the training data set (143) is greater than or equal to a predefined second boundary value, then the corresponding sensor data (173) is removed (311) from the training data set (143) or is not considered for the training data set (143).

3. The method (300) according to claim 1 or 2, wherein: The method further comprises: Generating (313) generated sensor data (191) for the training data set (143) by the classifier module (183), wherein the generating (313) includes: generating (315) by the classifier module (183) a latent space representation (189) having a predefined distance (D) in the latent space (195) to a preselected latent space representation (189), wherein the preselected latent space representation (189) is preselected classified sensor data (173); and generating (317) by the classifier module (183) sensor data (191) represented by the generated latent space representation (189), wherein the generated sensor data (191) has the preselected classification; and The generated sensor data (191) is added (319) to the training data set (143).

4. The method (300) according to any one of the preceding claims, wherein: The classification (307) also includes: The sensor data (173) of the training dataset (143) is labeled (321) corresponding to a classification.

5. The method (300) according to any one of the preceding claims, wherein: The method further comprises: The sensor data (173) of the training dataset (143) is visualized (323) by visualizing the corresponding latent space representation (189) of each sensor data (173) by a visualization module.

6. The method (300) according to any one of the preceding claims, wherein: The distance (D) is defined as the Euclidean distance.

7. The method (300) according to any one of the preceding claims, wherein: The classifier module (183) is constructed as a correspondingly trained artificial intelligence, which is configured to generate a corresponding latent space representation based on the sensor data (173) and / or to generate corresponding sensor data (173) based on the latent space representation (189).

8. The method (300) of claim 7, wherein: The classifier module (183) is constructed as a generative adversarial network (GAN), in particular as an autoencoder having an encoder module (185) and a decoder module (187).

9. The method (300) according to any one of the preceding claims, wherein: The received sensor data (173) includes tagged and / or untagged sensor data (173).

10. The method (300) according to any one of the preceding claims, wherein: The sensor data (173) includes radar data (103) of a radar sensor, and / or data of an inductive sensor and / or an eddy current sensor and / or a capacitive sensor and / or an AC current sensor and / or a nuclear magnetic resonance sensor and / or an ultrasonic sensor.

11. The method (300) according to any one of the preceding claims, wherein: The sensor data (173) are also classified with respect to an object depth (119) and / or an object extent (121) of the object (113).

12. The method (300) according to any one of the preceding claims, wherein: The object categories of the object type (117) of the object (113) include: metal / non-metal objects, cables for low voltage, cables with single-phase AC signals, cables with multi-phase AC signals, wooden brackets, metal brackets, plastic pipes, plastic pipes filled with water, such as tap water pipes, plastic pipes not filled with water, such as sewage pipes, and / or wherein the wall type categories of the wall type (123) of the wall (105) include: concrete walls, light structure / dry structure walls, brick walls and / or brick and stone walls, floor heating facilities, wall heating facilities.

13. A training data set (143) for training an artificial intelligence (125) of a measuring device (100) for wall diagnostics, wherein: The training data set is generated according to the method (300) for generating a training data set (143) according to any one of the preceding claims 1 to 12.

14. A computing unit (151) configured to implement a method (300) for generating a training data set (143) according to any one of claims 1 to 12 for training an artificial intelligence (125) for operating a measuring device (100).

15. A computer program product (500) comprising instructions which, when executed by a data processing unit, cause the data processing unit to implement a method (300) for generating a training data set (143) according to any one of claims 1 to 12, the training data set (143) being used to train an artificial intelligence (125) for operating a measuring device (100).

Images