Image processing apparatus, image evaluation apparatus, and manufacturing release system

Abstract

The present disclosure relates to an image processing device (100) for supporting a qualitative and / or quantitative evaluation of the quality of a crimp connection, an image evaluation device (200) having an image evaluation device (200) according to any one of claims 7 to 9, and a manufacturing release system (300) for a crimp device, the image processing device has a data interface connected to a database (304) in which a target value of the crimp connection (C) dependent on the production order is stored, the release unit (305) being designed to release the crimp connection (C) after a comparison between the at least one qualitative and / or quantitative quality parameter (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11) and the respective target value, and the release unit (305) being designed to release the crimp connection (C) from the database (304), after a comparison between the at least one qualitative and / or quantitative quality parameter (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11) and the respective target value. A release or rejection is provided to produce the classified crimp connection (C). This makes the determination of the quality parameters of the crimp connection more robust and reliable and makes the production of the corresponding crimp connection more reliable and less labor-consuming.

Description

Technical Field

[0001] The present disclosure relates to an image processing device for supporting qualitative and / or quantitative evaluation of the quality of crimp connectors. Further, the present invention relates to an image evaluation device for qualitative and / or quantitative evaluation of the quality of crimp connectors, and a manufacturing release system for a crimping device for producing crimp connectors. Background Art

[0002] In the production of cables, especially before or during the execution of a production order, it is necessary to ensure that the corresponding manufactured crimp connectors for the cables (such as data cables) are defect-free.

[0003] For this purpose, a so-called grinding image of the manufactured crimp joint is created before the start of production or during the production of a production order. For this purpose, the crimp connector is usually cut perpendicular to the longitudinal central axis, and specific quality parameters of the cross-section of the cut crimp connector are inspected. Thus, the grinding image is a cross-sectional image of the crimp connector.

[0004] The grinding image determines the crimp dimensions and helps to develop the crimp connector in terms of verifying the crimp quality of the crimping tool that is part of the crimping device.

[0005] For this purpose, the cross-section of the crimp connector is usually captured in digital and magnified form, for example using a microscope, to check for the presence of crimp defects.

[0006] It is known to use classical image evaluation algorithms to evaluate such cross-sectional images of crimp connectors to determine the corresponding crimp defects or quality parameters of the crimp connectors. Usually, the results of the classical image evaluation algorithm are manually inspected, and the results are manually corrected in case the results of the image evaluation software are inaccurate.

[0007] Alternatively, the crimp quality of the crimp connector is completely manually inspected based on the captured cross-sectional image of the crimp connection.

[0008] US2021295487A1 discloses a method for evaluating the crimp state of a crimp connector for a cable harness. An image of a part of the crimp connector of the cable harness is captured. Thereby, first data related to voids in the crimp connector part is determined. Then, based on these first data, the crimp state of the crimp connector part is determined. The cross-sectional image of the crimp connector is particularly used for this purpose, see Figure 2 、 Figure 4 、 Figure 6 。

[0009] CN111665267A discloses a visual recognition method for the crimp quality of a contact, which solves the problems that the crimp quality of the existing contact body and wire cannot be visually and depth-recognized and the crimp quality problems are often not detected.

[0010] EP3109624 A1 discloses a cable inspection system. The cable inspection system includes a mirror device having an odd number of sides arranged to form a pyramidal structure around a cable segment. A camera captures multiple images of the cable segment reflected by the mirror, each image showing a different side of the cable segment.

[0011] US2023245299A1 discloses a connector inspection system for a crimping machine. The connector inspection system includes an image processing device that takes pictures of a connector to be inspected and generates a digital image of the connector. The connector inspection system further includes a connector inspection module that communicates with the image processing device to receive the digital image of the connector as an input image. The connector inspection module has a reference image. The connector inspection module compares the input image with the reference image and performs semantic segmentation between the input image and the reference image to generate an output image. The output image shows the difference between the input image and the reference image to identify potential defects.

[0012] EP2173015A1 discloses a method for determining the crimp quality between a conductor and a contact, in which a crimping force is first applied to the conductor and the contact using a crimping device. From the crimping force curve generated during the crimping process, a normalized force-displacement crimping force curve is derived, and the compression area under the reference crimping force curve is determined. The crimping force curve and the reference crimping force curve are divided into multiple regions, and the division takes into account the size of the compression region. Another area under the crimping force curve is determined and used to convert the quality of the crimped connection.

[0013] US7174324B2 discloses a system in which, based on the input of known connection data, an estimation unit has pre-learned the relationship between the known connection data belonging to the connection structure and the unknown connection data belonging to the connection structure, and the estimation calculates the unknown connection data of the known connection data based on the learning result.

[0014] A computer vision system for automating the final inspection of crimped connections was disclosed by Giang Nguyen Huong et al. in "Deep learning-based automated optical inspection system for crimp connections" (XP033892793). The image processing chain for various defect classes and the model for analyzing the image data of crimped connections based on deep learning were described.

[0015] The required documents for the corresponding results of the crimp connection inspection, as part of quality assurance, are usually created manually, sometimes even handwritten, stored in a specified database, and archived.

[0016] Previous methods for determining the quality of crimp connections were expensive because they were labor-intensive and often led to non-reproducible results due to the manual and personal handling by the tester when determining quality parameters, especially quantitative quality parameters of the crimp connection.

[0017] The manual recording of the grinding image evaluation further complicates the traceability and retrievability of the archived results of the grinding image evaluation because these results are usually not machine-readable and thus not searchable. Summary of the Invention

[0018] The object of the present invention is to provide a solution by which the determination of the quality parameters of a crimp connection can become more robust and reliable, and by which the production of the corresponding crimp connection can become more reliable and less labor-intensive.

[0019] This object is achieved by the subject matter of the independent claims. Further improvements of the present disclosure are specified in the dependent claims, the description, and the drawings. In particular, the independent claims of one category can also be further improved analogously to the dependent claims of another category. More embodiments and improvements can be derived from the dependent claims, the description, and the reference drawings.

[0020] In particular, the present disclosure includes an image processing device for supporting the qualitative and / or quantitative evaluation of the quality of a crimp connection. The image processing device has a receiving unit for receiving a cross-sectional representation of an image, in particular, the cross-sectional representation is designed as a digital microscope image in the visible spectral range of the crimp connection. The image processing device has a processing unit designed to generate a raster image of the received image based on the received image using a deep neural network, wherein at least some pixels of a relevant image region of the received image can be assigned to a pre-determined category using the trained deep neural network, in particular, each pixel of the received image can be assigned to a pre-determined category. Furthermore, the processing unit is designed to generate at least one vector contour based on the generated raster image and generate and output an output signal based on the determined vector contour, and at least one qualitative and / or quantitative quality parameter assignable to the crimp connection can be determined based on the output signal.

[0021] Such an image processing device provides a robust and reliable means for processing cross-sectional images of crimp connections, enabling a reliable, objective, and reproducible evaluation of the images regarding qualitative and / or quantitative quality parameters to be performed with high reliability.

[0022] In particular, such an image processing device makes it possible to process defects during the image acquisition step so that they do not adversely affect subsequent image evaluation. For example, compensation can be made for inaccurate focusing, adverse lighting conditions, reflections on relevant parts of the cross-sectional image, or dirt on the ground surface, etc.

[0023] In particular, these defects may include cross-sectional contamination of the crimp connection, chamfering or abnormal burrs, which may lead to errors when evaluating images using classical image evaluation algorithms.

[0024] Classical image evaluation algorithms are understood as image evaluation methods that perform image evaluation independently of machine learning models.

[0025] Such defects in the grinding image particularly occur in a production-related environment. This results in limitations in the automation and accuracy of using traditional image evaluation algorithms to determine the quality parameters of crimp connections in a production-related environment.

[0026] This drawback can be eliminated by a corresponding image processing device because it is robust in overcoming corresponding interferences in image acquisition and thus can first achieve reliable automation.

[0027] Therefore, such an image processing device also allows for the execution of cost-effective, reliable and standardized methods for image evaluation.

[0028] Assigning at least some of the pixels in the relevant image region of the received image to a predetermined category, in particular assigning each pixel of the received image to a pre-determined category, can also be referred to as semantic segmentation of the received image. Thus, different categories form a raster image of the semantic segmentation of the crimp connection, which is usually a schematic representation of the cross-section of the crimp connection.

[0029] Through such semantic segmentation of the cross-sectional image of the crimp connection, in particular at least the relevant image region of the entire cross-sectional image, defects present in the original image that have a negative impact on the evaluation using traditional evaluation algorithms can be eliminated.

[0030] The relevant image region is the image region of the cross-sectional image of the crimp connection in which the crimp connection is depicted. The surrounding region within a pre-determined minimum distance around the outer boundary of the crimp connection does not necessarily need to be subjected to image processing and / or evaluation because it is usually irrelevant to the quality parameters of the crimp connection.

[0031] Preferably, the cross-sectional image is captured digitally, for example, using a CCD camera behind the microscope eyepiece to generate an image within the visible spectrum range of the human eye. It may be advantageous to expand the capture area of the image capture device in such a way that the cross-section of the crimp connection is completely depicted and at the same time has a magnification high enough to analyze the structure of the cross-section.

[0032] The quality parameters can include qualitative and quantitative quality parameters. Qualitative quality parameters can be, for example, statements as to whether the crimp connection or a specific quality parameter is acceptable, i.e., defective or non-defective according to the manufacturing specifications. Quantitative quality parameters can include quantitative characteristics that measurably characterize the crimp connection, such as crimp height, crimp width, measurable crimp width, support angle, support height, side end distance, crimp side end distance, burr height, burr width, bottom thickness, cavity between the wires, crack. The quality parameters of the crimp connection can be determined qualitatively and quantitatively.

[0033] The evaluation process is supported by a correspondingly designed image processing device. For example, based on the processed grinding image of the crimp connection, especially based on at least one determined vector contour, the evaluation process can continue using classical image evaluation algorithms.

[0034] The image processing device includes a receiving unit for receiving the cross-sectional image of the crimp connection, especially a photographic cross-sectional image of the crimp connection in digital form. This cross-sectional image advantageously but not necessarily shows the entire cross-section of the crimp connection, especially the maximum size within the range of the image capture device used. The image can be captured in such a way that the outer boundaries of the crimp sleeve in the radial direction are completely depicted.

[0035] The image capture device can optionally be included in the image processing device. For example, the image capture device can be a microscope mounted near the product. The cross-sectional image of the crimp connection captured by the image capture device can be provided to the receiving unit of the image processing device. Preferably, the image processing device is arranged in the monitoring area and connected to the image capture device such that the image captured by this image capture device can be provided to the receiving unit.

[0036] The received cross-sectional image of the crimp connection can be further processed, especially prepared, by a processing unit. For this purpose, the processing unit includes a trained deep neural network, using which the received image can be semantically segmented. That is, using this trained deep neural network, at least the pixels of the relevant image area, i.e., the area showing the cross-section of the crimp connection, especially each pixel of the received cross-sectional image of the crimp connection, can be assigned to a class.

[0037] In particular, through such semantic segmentation, relevant image content regarding quality parameters of the cross-sectional image of the crimp connection can be identified. The categories to which pixels can be assigned depend on the training data and the training of the deep neural network.

[0038] In this case, any deep neural network topology suitable for performing image semantic segmentation can be used.

[0039] Then, the thus segmented image can be converted from a raster image into a vector contour using a processing unit designed for this purpose. A vector contour is a graphical representation structure defined by the mathematical expressions of straight lines, curves, and shapes. Different from the raster view which consists of pixels and distorts when magnified, due to the mathematical expressions describing the relationships between various points, lines, and curves of the representable structure, vector graphics maintain their sharpness and quality at any size change. This improves the accuracy of subsequent evaluations, especially when determining at least one actual value of a quantitative quality parameter.

[0040] The conversion of the raster image to the vector contour is particularly advantageous in terms of the accuracy of subsequent evaluations. This means that subsequent evaluations are no longer limited by the number of pixels in the received image, thereby improving the accuracy of the determination regarding quality parameters, especially regarding the determination of quantitative quality parameters. The vector contour can be designed, for example, in the form of surfaces and / or lines.

[0041] The processing unit is designed to generate and output an output signal based on the vector contour, which enables the determination of qualitative and / or quantitative quality parameters assignable to the crimp connection. This can be the vector contour itself generated, which can subsequently be used for further image evaluations, for example, using classical image evaluation algorithms or manual evaluations. In particular, the vector contour can be output as a graphical representation.

[0042] The conversion of the resulting raster image to the vector contour with at least three categories can be performed conventionally, i.e., without using a trained deep neural network. Persons of ordinary skill in the art are familiar with the appropriate procedures for this purpose. Therefore, the output signal can also be generated and output outside of the trained deep neural network.

[0043] Alternatively, the trained deep neural network can also be designed to generate a vector contour based on the raster image. In addition, the trained deep neural network can also be designed to generate and output an output signal.

[0044] The output signal can be designed in such a way that the technical device provided with the output signal can process the output signal. In particular, the output signal can be designed such that the result of the image processing can be reproduced on an image display device.

[0045] However, the output signal can also be provided in such a way that it can be provided to an evaluation unit, based on which at least one qualitative and / or quantitative quality parameter can be determined, and at least one qualitative and / or quantitative quality parameter can be automatically determined by the evaluation unit.

[0046] In another embodiment, the trained deep neural network includes an image transformer network, also known as a vision transformer or ViT, and / or a convolutional network, also known as a convolutional neural network or CNN. These are deep neural networks particularly suitable for the intended image processing.

[0047] In particular, the trained deep neural network can be designed as an image transformer network or a convolutional network. Using an image transformer network is particularly advantageous because it requires much less training effort compared to other suitable network topologies. The image transformer network is an encoder-decoder network suitable for image classification.

[0048] The trained deep neural network, particularly a deep convolutional network, is trained using an appropriate training data set and is designed for semantic segmentation of cross-sectional images of crimped connectors.

[0049] As part of such training, corresponding annotated training data can be presented to the trainable neural network, which includes the desired classes that the trained network will subsequently identify.

[0050] For example, such a training data set can be manually created from already available historical data, such as from defect-free and defective grinding images based on past production orders. Cross-sectional images of such crimped connectors are readily available because cable production is routinely monitored, so grinding images are made at regular intervals, and for quality assurance reasons, this data is available over a relatively long period of time.

[0051] The training of the deep neural convolutional network is described as follows. It should be understood that the creation and training of the described training data form an independent subject.

[0052] For the training of the deep neural network, a corresponding set of training images is generated. These images include cross-sectional images of crimped connectors and the desired class assignments for the pixels.

[0053] For example, all pixels showing the inner conductor are assigned to one class, all pixels showing the crimp sleeve are assigned to another class, and all pixels located radially outside the crimp sleeve are assigned to yet another class.

[0054] The training data here includes not only defect-free crimped connectors, but also defective crimped connectors, such that these pixels can be assigned to multiple classes regardless of their relative arrangement on the cross-sectional image.

[0055] The capture of the corresponding images and their identification can be carried out during normal production or test operations, so that a variety of images with different crimps, sizes, viewing directions, rotations, contrasts, illuminations, obstacles, etc. can be generated, and multiple classes determined offline can be supplemented. The identification of such images can be done manually, or at least partially performed or supported by a traditional image processing system while manual rework may be required. Then the prequalified training images can be further preprocessed for training.

[0056] For example, as part of further preprocessing, the size of the training images can be adjusted to a predetermined size. In particular, the number of pixels in the training images can be adapted to the input number of the input layer of the neural network. The preprocessing of the training images can also include normalizing these images.

[0057] In one embodiment, the foregoing preparation and preprocessing of the images are not performed, enabling the trained neural network to process the corresponding raw data, which increases the subsequent speed of semantic segmentation using the trained deep neural network.

[0058] To improve the results of the neural network and make it more robust, random image processing can be performed on the training images. Such image processing can include, for example, rotation, magnification, reduction, and / or distortion. It should be understood that the corresponding amplitudes of each image processing can be specified.

[0059] For example, for magnification or reduction, the maximum magnification or reduction percentage can be specified, such as 110% or 90%. For rotation, the maximum or minimum rotation angle can be specified, such as + / −10°, 20° or 30°. Similarly, corresponding limits can be specified for distortion. It should be understood that different image distortion algorithms can be used, and these algorithms can have different parameters.

[0060] The image processing is used to provide greater variability to the training data. The expected multiple classes will thus appear at different positions in the image in different sizes. For example, this prevents the neural network from learning to recognize a given feature only in a small part of the image and wrongly concluding that the feature does not exist, even if the feature is, for example, only located outside that part.

[0061] After preprocessing the training images, a neural network is trained using a part of the trained images obtained in this way. By feeding the remaining training images to the trained neural network and comparing its output with the known or expected output of the corresponding training images, the remaining training images can be used to check whether the learning was successful. Thus, this part of the training images can also be referred to as test data.

[0062] After training is completed, the quality of the training can be checked as described above by the trained neural network using the validation of the test data. If the results meet the required quality, the training can be completed. If the results do not meet the required quality, or if the neural network is to be further trained, training can be continued with the corresponding training data or repeated with changed parameters.

[0063] It should be understood that depending on the type of neural network used, training can be performed in different ways.

[0064] Typically, the weights of the neural network are adjusted in each training run to minimize the error of the output of the neural network compared to the known results from the training dataset. This is usually done by means of so-called backpropagation, error feedback or backpropagation.

[0065] For training, an epoch number and a termination criterion can also be specified. The epoch number indicates the number of training runs. For each training run, a pre-determined amount of training data can be used, for example, all training data or only selected training data. The termination criterion indicates how far the results of the trainable neural network can deviate from the ideal results for the training to be considered successfully completed, and thus for the neural network to be sufficiently well trained.

[0066] Deep convolutional networks, in particular deep convolutional neural networks, also known as dCNNs, provide good to very good results, especially for classifying objects in image data. It should be understood that other suitable neural networks from the field of machine learning are also possible.

[0067] Such a dCNN can have an input layer, multiple hidden layers and an output layer. The hidden layers can at least partly comprise identical or repeated layers.

[0068] The input layer can have an input for each pixel of the captured image. It should be understood that the image can be transmitted to the input layer, for example, as an array or vector with a corresponding number of elements. In addition, for all images, the size of the captured image can be the same. For example, cross-sectional images can be captured with 1024×1024 pixels, covering a capture area of 2cm×2cm, 1cm×1cm or 2.5cm×2.5cm.

[0069] The capture region can be selected according to the crimping connector to be analyzed. Preferably, the capture region is selected such that the cross-section of the crimping connector is fully captured.

[0070] It should be understood that these specifications are merely examples and that other image parameters can be used.

[0071] For semantic image segmentation, there are some simple convolutional network architectures that can serve as the basis for more complex models or be suitable for simpler applications.

[0072] For example, the so-called Fully Convolutional Network (FCN) developed for semantic segmentation can be used. It consists of a series of convolutional layers that have been modified for image classification to generate pixel-accurate outputs.

[0073] Unlike classical convolutional neural networks (abbreviated as CNN) that are typically used to classify entire images, the FCN replaces the fully connected layers at the end of the network with convolutional operations to generate pixel-accurate outputs. The FCN also uses upsampling operations to scale the output to the original image size.

[0074] The FCN can be trained to integrate information from different scales to consider context and detail information for accurate segmentation. This can be done through different layers or modules operating at different scales.

[0075] The output layer of the FCN generates a probability distribution for each class for each pixel in the input image. This output represents the class to which each pixel is predicted, thus achieving pixel-accurate segmentation.

[0076] FCNs can be flexibly used and trained for various input image sizes. They are effective for semantic segmentation because they can generate accurate pixel-to-pixel assignments for different classes or objects in an image.

[0077] In an embodiment, an image transformer network can be used to perform semantic segmentation of the cross-sectional image of the crimping connector.

[0078] In the context of the present application, an image transformer network is understood as a neural network based on the transformer architecture, i.e., including an encoder, a decoder, or both. These are commonly referred to as Vision Transformers (ViT). These are transformer networks specifically developed for processing images.

[0079] For example, the image transformer network includes the following components. First, the network may include a unit for patch processing, in particular a unit for patch embedding. A patch is understood to be a part of the image to be processed. The patch processing unit divides the received image into a plurality of patches, i.e., a plurality of sub-images or a plurality of image parts, and treats each patch as a token used as an input to a subsequent transformer. In particular, a vector representation of the input data may be provided. In addition, the patch processing unit retains the position information of these sub-images.

[0080] The image or sub-image is thus converted into a digital representation. These patches are then treated as a sequence of tokens, similar to words in a sentence in natural language processing. Typically, the patches or sub-images overlap to better capture the local and global relationships of the image features.

[0081] In addition, the image transformer network typically includes a plurality of transformer blocks. These include so-called attention mechanisms, such as self-attention, attention matrices, or mechanisms for modeling the relationships between tokens or patches with each other. These transformer blocks process the sequence of patches to extract global and local features in the image and to model the relationships between the patches.

[0082] In self-attention, also known as self-attention, the relationships between each patch or token and other patches or tokens are captured and provided to the model.

[0083] For each patch or token, the attention mechanism calculates the weights or attention distribution of all other tokens or patches in the image. This weight indicates the degree of relevance of other parts of the image to the current token or patch.

[0084] The attention matrices show how each token or patch responds to other tokens or patches. These attention matrices indicate which parts of the image are considered more important and which parts are considered less important.

[0085] By capturing these relationships between tokens in the form of self-attention and attention matrices, the model can absorb context information and capture relevant visual relationships within the image. This allows for the effective processing of global and local context information.

[0086] The transformer block processes a sequence of patches to extract global and local features based on the attention mechanism described above and to model the relationships between patches. Additionally, these attention mechanisms are combined with a feedforward layer to update and refine the representation of the patches.

[0087] Additionally, an image transformer network is typically designed to preserve the spatial location information of the patches in an image, e.g., by inserting positional encodings or other mechanisms, particularly in the patch embedding area, such that statements provided by the image transformer network can be correctly reproduced in position.

[0088] Advantages of such an image transformer network include its good scalability, allowing it to be applied to images of different sizes. The network does not require a fixed input size as many CNN-based methods do. The transformer mechanism allows the image transformer network to capture and utilize the global context information of the entire image.

[0089] Additionally, an image transformer network can be easily adapted to different tasks and tuned for various tasks, such as classification, object detection, image segmentation, etc. In particular, this can be achieved through a multi-layer perceptron downstream of the encoder included in the image transformer network.

[0090] In particular, an image transformer network can be trained in two stages. First, it can be pre-trained based on a large dataset to learn the task of semantic segmentation. Additionally, the image transformer network can be fine-tuned, e.g., by the user of the image transformer network using application-specific data for training.

[0091] The image transformer network also allows for parallel processing of data, which reduces the resources required for inference.

[0092] In particular, a SegFormer network, which is designed as an image transformer network, can be provided for semantic segmentation of cross-sectional images of crimp connectors. However, other transformer architectures for semantic segmentation can also be used, such as the Detection Transformer (DETR), Vision and Language Bidirectional Encoder Representations from Transformers (VilBERT), etc.

[0093] The training of an image transformer network, particularly the SegFormer network, can be accomplished similar to the procedure for training a CNN described above.

[0094] In an exemplary embodiment, the SegFormer architecture can be selected as follows to achieve robust and reliable results in image processing. However, the architectures described below are merely one possible design, and those skilled in the art are not limited to this design.

[0095] The SegFormer network, as an image transformer network, includes an encoder side and a decoder side like a transformer network.

[0096] For example, on the encoder side, four transformer blocks connected in series can be provided, and each transformer block is preceded by a unit for patch processing.

[0097] Each transformer block has an output through which data is provided, particularly the image data processed in the corresponding transformer block or a representation of the image data. The first transformer block among a plurality of connected transformer blocks processes the image with a relatively small number of patches, that is, the original image is divided into a relatively small number of usually overlapping patches, such as four patches (2×2).

[0098] Each patch is processed within the transformer block according to the above-mentioned attention mechanism and fed to a feed-forward network. Then, in a subsequent step, for example, new image data is generated based on this image data using overlap patch merging, and the new image data is provided to the output of the first transformer block.

[0099] Then, these output image data of the first transformer block are provided to a subsequent patch processing unit, which then provides a larger number of usually overlapping patches, such as 16 patches (4×4), to the subsequent second transformer block among a plurality of connected transformer blocks. Then, these patches are processed in the second transformer block and provided to the output of the second transformer block.

[0100] The image data at the output of the second transformer block among the four connected transformer blocks are then provided to a subsequent patch processing unit, which then provides, for example, 64 usually overlapping patches (8×8) to the subsequent third transformer block. These patches are processed in the third transformer block and provided to the output of the third transformer block.

[0101] The image data at the output of the third transformer block among the four connected transformer blocks are then provided to a subsequent patch processing unit, which then provides, for example, 256 usually overlapping patches (16×16) to the subsequent fourth transformer block. These patches are processed in the fourth transformer block and provided to the output of the fourth transformer block.

[0102] The image data at the output of the fourth transformer block is then provided to the subsequent tile processing unit and then fed into a multi-layer perceptron, abbreviated as MLP. In addition, all output data from each transformer block (i.e., all four transformer blocks, exemplarily) is fed into the multi-layer perceptron.

[0103] A multilayer perceptron (MLP) refers to a specific type of network layer used in the SegFormer architecture. A multilayer perceptron is a simple form of neural network that consists of at least three layers, and often more than three layers in applications.

[0104] There is an input layer that receives data, such as data from these transformer blocks. This layer represents the features of the input and passes that feature to the next layer, which is at least one hidden layer.

[0105] There are typically multiple hidden layers, but at least one hidden layer is located downstream of the input layer. This at least one hidden layer is called "hidden" because it is located between the input and output layers and is not directly visible from the outside. In a multilayer perceptron, there can be multiple hidden layers, and in particular, multiple hidden layers. Data fed to this at least one hidden layer is processed in a desired manner using weights created through training.

[0106] Furthermore, there is an output layer of the multilayer perceptron, in which the processed data is output and is typically further processed in a decoder area.

[0107] The decoder region of the SegFormer network typically has the task of scaling the identified relations to the original image size using previous blocks and modules, also known as upsampling, to obtain a segmentation result that is as pixel-accurate as possible.

[0108] The decoder area includes upsampling layers and corresponding transforms. For upsampling, bilinear upsampling can be used, for example, to scale the result to the desired size. This creates a smooth upscaling of the predictions.

[0109] The decoder integrates this upscaling with the local and global information captured by the transformer blocks in the encoder region. In this example, skip connections are also used to include detailed features from earlier layers in the processing and decoding of the multilayer perceptron. The decoder can also include an attention mechanism applied to the data being processed.

[0110] The output of the decoder of the image transformer network is a reconstructed segmentation mask that represents the predicted class or category for each pixel in the image.

[0111] Like an encoder, a decoder can be implemented in different ways. Examples of decoder implementations can be obtained from relevant platforms such as huggingface.co or github.

[0112] However, the function of the decoder is generally intended to bring the magnified result or prediction to the original image size and provide a detailed pixel - precise segmentation result.

[0113] In another embodiment, the deep neural network is designed as an image transformer network, including a publicly available pre - trained image transformer network and at least one network layer, particularly an additional network layer subsequently added by the user, especially an additional network layer specifically trained subsequently using application - specific training data for image processing of cross - sectional images of crimp connectors.

[0114] Application - specific training data is understood as data provided from the specific application for which the image transformer network will be used, here specifically, for example, for the application of segmentation and / or evaluation of cross - sectional images of crimp connectors. For example, suitable pre - trained image transformer networks for semantic segmentation can be obtained via huggingface.co or github.

[0115] The pre - trained image transformer network has the advantage of being easily adaptable and only requires a small amount of additional training work to customize them for a specific task or application. Therefore, it is possible to use publicly available pre - trained image transformer networks, and it is easy for the user to adapt them to specific applications subsequently.

[0116] The subsequent adjustment of the pre - trained model to a specific application can be carried out by adding at least one new layer or by appropriately training at least one existing layer of the model. Then, training with application - specific data leads to a corresponding adaptation of the model based on this training data, thus generating improved results.

[0117] On the one hand, for a specific application regarding cross - sectional images of crimp connectors, a simple adaptation of the pre - trained image transformer network can generally be carried out.

[0118] In addition, the nature of the crimp connector can vary depending on the crimping device. Therefore, the application - specific data may be specific to the crimping device or the crimping tool. This means that due to the small amount of training work, the image transformer network can be individually trained for each crimping device or each crimping tool, thus providing further improved results in terms of image processing and subsequent image evaluation.

[0119] In another embodiment of the image processing device, the trained deep neural network is designed to generate a raster image with at least three classes based on the received image. The first class corresponds to the internal components of the crimp connection, in particular the conductor elements. The second class corresponds to the external components of the crimp connection, in particular the crimp sleeve. The third class corresponds to the environment of the crimp connection.

[0120] It has been shown that providing three classes is sufficient for robust and reliable image processing. However, additional classes can also be provided within the scope of image segmentation.

[0121] The first class corresponds to the internal components of the crimp connection and does not necessarily have to be designed as conductor elements, for example in the form of wires or solid internal conductors, but can also be of a more complex configuration. For example, in the case of a sheath crimp, the cable configuration is surrounded by a crimp sleeve and includes the cable sheath.

[0122] The second class of the at least three classes can correspond to the external components of the crimp connection. In particular, this class can be selected to correspond to the crimp sleeve of the crimp connection. In particular, this can be a crimp sleeve for sheath crimping or internal conductor crimping.

[0123] The third class of the at least three classes can correspond to the environment of the crimp connection. The environment is not part of the crimp connection. The environment can be air or other mechanical components, such as components that guide the crimp connection during image capture or other structures of the cable that are not considered in the analysis.

[0124] By providing these three classes within the scope of image processing, subsequent image evaluation can already be significantly improved. In particular, by accurately classifying the internal components of the crimp connection, the crimp sleeve, and its demarcation from the environment, a robust, reliable, and accurate basis can be provided for subsequent evaluation.

[0125] The at least three classes can be assigned corresponding different values. In particular, these values can be any distinguishable value ranges, where each class is assigned a value range. The corresponding value range can also be a single value for one class. The assignment of pixels to the corresponding values can be based on probabilities. A pixel is given the value that is most likely to describe the pixel.

[0126] In one embodiment of the image processing device, the trained deep neural network is designed to assign colors to each of the first, second, and third classes, where adjacent classes can be displayed as regions of different colors, especially colors with a distinct contrast, in the raster image and / or vector contour, and the color assignment is included in the output signal.

[0127] This embodiment is a suitable basis for an artificial assessor and traditional image assessment software to determine at least one qualitative and / or quantitative quality parameter of the crimp connector.

[0128] Through different color designs of at least a first category, a second category, and a third category, the contours and extents of the corresponding regions of the crimp connector, that is, for example, the internal components, external components, and the environment of the crimp connector, can be quickly and clearly captured by the artificial assessor and / or the image assessment software, significantly increasing the reliability of the assessment.

[0129] This is particularly simple if the colors of adjacent categories provide a good enough contrast such that the image assessment software and / or the artificial assessor can easily distinguish between them.

[0130] In particular, this allows the boundary contours between adjacent color regions, that is, the boundary lines between the first category and the second category or between the second category and the third category, to be well identified.

[0131] In particular, exactly three categories can be provided, including the first, second, and third categories.

[0132] The output signal can be designed such that the cross-sectional image of the crimp connector can be displayed as a three-color image with different colors. This is a particularly simple and clear representation of the cross-sectional image of the crimp connector for reliably assessing its quality parameters manually or with the help of image assessment software.

[0133] In another embodiment, a trained deep neural network is designed to determine the boundary contours between the first category and the second category and / or between the second category and the third category. In particular, the boundary contours can be designed as line segments with a pre-determined thickness, which include or represent the boundary regions between adjacent categories or colors. The trained deep neural network can also be designed to generate an output signal using which the determined boundary contours can be graphically displayed.

[0134] For example, during the assessment using classical image algorithms, at least one boundary contour can be approximated as a broken line, and the qualitative and / or quantitative quality parameters of the crimp connector can be determined based on this.

[0135] For example, the cross-sectional image of the crimp connector can be used to train such a network to determine the boundary contours, where the corresponding boundary contours have been manually added. The training data can be increased and changed in the usual way, especially as described above, to improve network training.

[0136] The image processing device can be designed to specifically identify at least one point of the boundary profile, based on which the determination of specific quantitative quality parameters will be performed. For example, to determine at least one of the following quality parameters: crimp height, crimp width, measurable crimp width, support angle, support height, side end distance, crimp side end distance, burr height, burr width, bottom thickness, cavity between wires, crack.

[0137] These points are called marker points because they mark reference points for possible subsequent quality parameter measurements. In particular, marker points can also be provided to mark defect structures in the cross-sectional image of the crimp connection, such as the cavity between wires. If present, such a cavity will appear in the internal area of the crimp connection, corresponding to the first category.

[0138] Therefore, these marker points can mark the points of the boundary profile that are necessary for determining the required quality parameters and are thus particularly necessary for the quantitative evaluation of the cross-sectional image.

[0139] For example, simple logical operations, classical image evaluation algorithms, or deep neural networks trained for this purpose can be used to determine the marker points.

[0140] The boundary profile, especially the line segments representing the boundary profile, can preferably be superimposed on the received cross-sectional image of the crimp connection. This creates the possibility for a sanity check because the boundary profile is visually displayed in the received cross-sectional image of the crimp connection. Based on the boundary profile, the quantitative and qualitative quality parameters of the crimp connection can be determined accordingly. In particular, the relevant marker points of the corresponding quality parameters of the crimp connection can also be displayed.

[0141] In another embodiment of the image processing device, the output signal is designed to include a superimposed display based on the received image and the vector profile, especially at least one boundary profile, where the measurement points for determining the corresponding quality parameters are included as separately visible marker points in this superimposed display. This makes the evaluation based on this display particularly easy. In particular, the marker points can be designed to be distinguishable, for example, in different colors, so that they can be directly assigned to specific, especially quantitative, quality parameters. This helps to perform a reliable and error-free evaluation. The marker points can be determined using corresponding logical operations based on the vector profile.

[0142] The present disclosure also particularly includes an image evaluation device for qualitative and / or quantitative evaluation of the crimping quality. The image evaluation device includes the image processing device according to any one of claims 1 to 6, and determines a predetermined qualitative and / or quantitative quality parameter of the crimping connection by using an executable error classification based on the generated vector contour. The image evaluation device is designed to generate and output an evaluation output signal according to the executed error classification. In particular, the error classification includes the executed error classification for determining the predetermined qualitative and / or quantitative quality parameter of the crimping connection.

[0143] Therefore, the image evaluation device is not only used for image processing, but also for image evaluation of the cross-sectional image of the crimping connection. This allows for reliable, objective, and robust determination of the qualitative and / or quantitative quality parameters of the crimping connection, because manual evaluation can be omitted.

[0144] The error classification for determining the predetermined qualitative and / or quantitative quality parameter of the crimping connection can be performed using classical image evaluation software that is not based on a neural network or otherwise. In particular, this can be done using a separate evaluation unit, to which the output signal of the image processing device can be provided.

[0145] The term error classification will be understood in a broad sense in the context of the present disclosure. Therefore, error classification not only relates to the presence of errors in the cross-sectional view of the crimping connection, but also to the absence of errors. The error classification can essentially be qualitative, for example, the crimping connection is "acceptable" or the crimping connection is "unacceptable". The error classification can also be designed to qualitatively and / or quantitatively classify different error images. Here, quantitative determination is also understood as classification. Therefore, the error classification particularly includes the classification of the qualitative and quantitative quality parameters of the crimping connection represented as a grinding image, as well as the measurement of quantitative quality parameters in the form of actual values from the vector contour, for example.

[0146] In the context of image evaluation, an actual value can be assigned to the quantitative quality parameter, and this actual value is understood as the measured value of the corresponding quality parameter. This can be compared with the predetermined target value of the corresponding quality parameter.

[0147] To determine the actual value, in addition to the cross-sectional image of the crimping connection, the image evaluation device can also be provided with a reference scale, thereby allowing conversion of image features into, for example, SI units or other suitable scales, such as a test bench-specific conversion value from pixels to millimeters. This makes it possible to easily compare with possible target values. For example, the target value can be specified in SI units.

[0148] In one embodiment, the image evaluation device includes a trained deep neural network, and by using this trained deep neural network, misclassification for determining predetermined qualitative and / or quantitative quality parameters of a crimped connection can be performed.

[0149] By using the trained deep neural network, misclassification can be effectively and reliably performed. The image evaluation device may in particular include a separately trained deep neural network, which is designed to perform an evaluation based on the output signal of an image processing device. Thus, the image evaluation device may include two independent deep neural networks connected in series, with the first trained deep neural network designed to perform image processing and the second trained deep neural network designed to perform image evaluation based on the output signal of this image processing.

[0150] The deep neural network is trained for the corresponding classification task by using training data reflecting the corresponding misclassification in the above-described manner. Misclassification for determining predetermined qualitative and / or quantitative quality parameters of a crimped connection can be performed, in particular by using a trained image transformer network or a trained deep convolutional network.

[0151] In another embodiment, the image evaluation device is designed such that image processing and misclassification for determining predetermined qualitative and / or quantitative quality parameters of a crimped connection can be performed by using a common deep neural network. Thus, image processing and image evaluation can be combined. The image processing as the basis for evaluating a ground image of a crimped connection is directly related to image evaluation.

[0152] This has the advantage that only one training process for image processing and subsequent misclassification is required, while providing a robust, reliable and objective evaluation of the quality parameters of the crimped connection.

[0153] If the pre-trained image transformer network is used for semantic segmentation, this can be easily extended to misclassification and determining the actual value of the quantitative quality parameter.

[0154] In another embodiment, the evaluation output signal includes at least one quality parameter from the following: defect-free crimped connection, crimp height, crimp width, measurable crimp width, support angle, support height, side end distance, crimp side end distance, burr height, burr width, bottom thickness, cavity between wires. In particular, the evaluation output signal may include at least one measured value or actual value of at least one of the foregoing quality parameters.

[0155] It is based on error classification for at least one of the following error categories: defect-free crimp connectors, crimp height, crimp width, measurable crimp width, support angle, support height, side end distance, crimp side end distance, burr height, burr width, bottom thickness, cavity between wires. It is preferable to use all these quality parameters, which can comprehensively characterize the corresponding crimp connectors.

[0156] The following quality parameters can be determined quantitatively and qualitatively: crimp height, crimp width, measurable crimp width, support angle, support height, side end distance, crimp side end distance, burr height, burr width, bottom thickness, cavity between wires.

[0157] In another embodiment of the image evaluation device, the evaluation output signal is designed to include a superimposed display based on the received image and vector contours, especially at least one boundary contour, wherein the measurement points for determining the corresponding quality parameters are included in the superimposed display as separately marked marker points, especially separately visible marker points.

[0158] Marker points are understood as markers in the image that mark specific points in the image assigned to the corresponding quality parameters. In particular, using these markers, for example, located on the edges of the boundary contours, distances can be calculated, for example, in pixels. These distances can be converted into length units using a conversion ratio.

[0159] In an embodiment where the marker points are used as reference points for determining the actual values of the crimp connectors, the marker points do not have to be perceptible to the user. If there are marker points for the quality parameters of the crimp connectors and they can be used to determine the actual values, it is usually sufficient to determine the actual values of these parameters.

[0160] However, it is advantageous if the marker points are separately marked so that the position of the marker points can be quickly and easily identified by the observer of the superimposed display. This allows the monitoring personnel to conduct a quick, possibly spot-check review of the automatic evaluation. This can be carried out especially to check whether the image evaluation device is working error-free.

[0161] In particular, different marker points can be used for different quality parameters so that the different marker points can be distinguished from each other. In particular, a trained deep neural network can be designed to locate and mark the relevant marker points in the image. In addition, the trained deep neural network can be designed to generate an output signal using which the relevant marker points can be displayed in the image.

[0162] Different marking points can be used for one or more of the following quality parameters: defect-free crimp connections, crimp height, crimp width, measurable crimp width, support angle, support height, side end distance, crimp side end distance, burr height, burr width, bottom thickness, cavity between wires, and cracks. In this way, each marking point can be easily assigned as a reference point for relevant measurements.

[0163] In another embodiment, the image evaluation device is designed to provide an evaluation output signal to a central database, in particular a manufacturing database. In particular, this can be done by using a data interface included in the image analysis device.

[0164] This allows the determined quality parameters to be stored or archived and recorded in a traceable manner. This enables the results of the image evaluation to be automatically assigned, recorded, retrieved, and traced back to the corresponding manufactured crimp joints. By eliminating the manual recording of qualitative and / or quantitative quality parameters of the crimp connections, the consistency of the stored data can be ensured, which helps to improve data evaluation and management.

[0165] In addition, the image evaluation device can include an evaluation unit in which the quantitatively determined quality parameters are compared with the target values of the manufactured crimp joints. Depending on the result of the comparison, a control signal can be generated to affect the usability of similar crimp connections and / or affect the production of other similar crimp connections.

[0166] The target values can be provided to the image evaluation device, in particular to the evaluation unit of the image evaluation device, by using a central database, especially a manufacturing database, which is technically data-connected to the image evaluation device.

[0167] In the connectable central database, especially in the manufacturing database, the corresponding target values of the corresponding production orders of the crimp joints can be stored. In the case of being connected to the image evaluation device, these target values can be sent or requested or received by the image evaluation device according to the corresponding production orders of the image evaluation device for comparison with the determined actual values.

[0168] The present disclosure also relates to a manufacturing release system for a crimping device having an image evaluation device according to any one of claims 7 to 12. The system has a data interface connected to a database in which target values dependent on production orders for crimp connections are stored. The system has a release unit designed to release or reject the release of the manufactured classified crimp connections after comparing at least one qualitative and / or quantitative quality parameter of the crimp connections with the corresponding target values.

[0169] Thus, a comparison is made based on an evaluation output signal provided by an image evaluation device, the evaluation output signal including at least one qualitative and / or quantitative quality parameter. Based on this comparison, production is released or not released.

[0170] In another embodiment, the manufacturing release system includes a release unit configured to provide release or non-release of a classified delivery of a press connection based on the evaluation output signal.

[0171] Thus, the release unit is configured to release not only the manufacturing but also the delivery of the crimped connection. This may be necessary if manufacturing defects are only detected at a later stage, for example by means of corresponding samples. In particular, a delivery stop can be recorded in a central database, thus preventing defective crimped connections from being delivered to the customer.

[0172] In another embodiment, the manufacturing release system includes a database, wherein the evaluation output signal includes at least one storable quality parameter from among: defect-free crimped connection, crimp height, crimp width, measurable crimp width, support angle, support height, side end distance, crimp side end distance, burr height, burr width, bottom thickness, cavity between conductors, crack.

[0173] The image processing device, receiving unit, processing unit, image evaluation device, image evaluation unit, comparison unit, database, release unit or any other unit described herein may include or may be provided in at least one of or as part of a dedicated processing element (e.g., a processing unit, microcontroller, field programmable gate array, FPGA, complex programmable logic device, CPLD, application specific integrated circuit, ASIC, etc.). A corresponding program or configuration may be provided to implement the required functions. The image processing device, receiving unit, processing unit, image evaluation device, image evaluation unit, comparison unit, database, release unit or any other unit described herein may also be at least partially provided as a non-transitory computer program product including computer-readable instructions executable by a processing element. In another embodiment, the image processing device, receiving unit, processing unit, image evaluation device, image evaluation unit, comparison unit, database, release unit or any other unit described herein may be provided as an additional or supplementary function or method of the firmware or operating system of a processing element that already exists as corresponding computer-readable instructions in a corresponding application. Such computer-readable instructions may be stored in a memory coupled to or integrated into the processing element. The processing element may load the computer-readable instructions from the memory and execute them.

[0174] Furthermore, it should be understood that any required support or additional hardware may be provided, such as a power supply circuit and a clock generation circuit.

[0175] Generally, any computer program or computer program product disclosed herein should be understood as a non - transitory computer program product. BRIEF DESCRIPTION OF THE DRAWINGS

[0176] The following embodiments of the present disclosure are explained with reference to the accompanying drawings. These drawings show:

[0177] Figure 1 : A schematic diagram of an embodiment of an image processing device.

[0178] Figure 2 : A schematic diagram of the structure of an exemplary image transformer network configured as a SegFormer.

[0179] Figure 3 : A schematic representation of a received image of a 3 - color raster image segmented semantically and an overlapping image of the received image and the determined boundary contour, in particular, can be generated by an image processing device according to Figure 1 thereof.

[0180] Figure 4 : A schematic diagram of a first embodiment of an image evaluation device.

[0181] Figure 5 : A schematic diagram of a second embodiment of an image evaluation device.

[0182] Figure 6 : A schematic representation of a received image of a 3 - color raster image segmented semantically, an overlapping image of the overlapping image of the received image and the determined boundary contour, and a quantitative quality parameter that can be generated by an image evaluation device according to Figure 4 or Figure 5 thereof.

[0183] Figure 7 : A schematic embodiment of a manufacturing release device.

[0184] Figure 8 : A schematic cross - sectional view of a crimp connection.

[0185] The drawings are only schematic representations and are only used to explain the present disclosure. The same or equivalent elements are always denoted by the same reference numerals. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0186] Figure 1 A schematic diagram of an image processing device 100 is shown. The device includes a receiving unit 101 for receiving a recorded cross - sectional image of a crimp connection.

[0187] This can in particular be a digitally recorded cross-sectional image of a crimped connection, in particular a microscopic image of a cross-section of a crimped connection. The plane of the cross-sectional image of the crimped connection is in particular perpendicular to the longitudinal central axis of the cable on which the crimped connection is arranged. However, if it seems helpful for determining the quality parameters of the crimped connection, another cross-section can be selected.

[0188] Such a cross-sectional image of the crimped connection is typically generated by cutting the crimped connection perpendicular to the longitudinal axis of the cable. The cross-sectional image is typically prepared by grinding. Therefore, such a cross-sectional image is also referred to as a ground image of the crimped connection.

[0189] The image processing device 100 also includes a processing unit 102. The processing unit 102 is used to process the received cross-sectional image. By processing the received image, the evaluation of the image can be improved because the processing can eliminate unwanted deviations from an ideal cross-sectional image with optimal recording conditions (which often occurs in practice).

[0190] For this purpose, the processing unit 102 includes a trained deep neural network, and the trained deep neural network is configured as an image transformer network. The image transformer network is trained to process the image in a desired manner.

[0191] In particular, the image transformer network is configured as a SegFormer and is designed to generate a semantically segmented raster image based on the received image. In this process, each pixel of the region of interest in the received image, in particular each pixel of the received image, can be assigned to a pre-determined class. The assignment is based on training data in which the corresponding classes are defined by corresponding annotations.

[0192] Advantageously, a publicly available pre-trained image transformer network is used and, in the form of fine-tuning, i.e., in the form of subsequent training suitable for a specific task, it is individually adapted to the segmentation task of the cross-sectional image of the crimped connection. This significantly reduces the training effort of the user of the image transformer network. For this purpose, for example, a pre-trained SegFormer can be used, and then the user of the pre-trained SegFormer network can further train it using appropriately annotated training data.

[0193] Based on the raster image generated by the SegFormer, the processing unit 102 can generate a vector contour. This means that the generated raster image is converted into a vector contour. This increases the accuracy of the subsequent evaluation of the processed image. Based on this, the processing unit 102 then generates an output signal based on which a subsequent improved evaluation can be performed.

[0194] Through this image processing, artifacts and inaccuracies can be eliminated, thereby avoiding interference of such artifacts and inaccuracies with the evaluation.

[0195] The output signal is configured to include at least one qualitative and / or quantitative quality parameter assignable to the crimp connection such that the parameter can be determined from the output signal. The at least one quality parameter can then be determined by conventional image evaluation software and / or manually.

[0196] In a first variant, converting the raster image to be generated into a vector contour can be done in a conventional manner, i.e., without using a neural network. For example, known image processing techniques and algorithms can be used. A common method is to apply the "Canny edge detector" technique in combination with the Hough transform.

[0197] However, not only can the raster image be generated by the image transformer network, but also, after corresponding training, the raster image can be generated by the vector contour itself, and if necessary, by the corresponding output signal. In this case, the image transformer network can also include a residual network for generating a vector contour based on the raster image.

[0198] Thus, after receiving a cross-sectional image of the crimp connection, all processing steps can be performed by a correspondingly trained deep neural network, in particular an image transformer network.

[0199] The raster view generated by the trained deep neural network typically includes image regions assigned to specific classes. In a simple form, this can be a first class corresponding to the internal region of the crimp connection, for example a first class corresponding to the internal conductor region. In addition, a second class corresponding to the crimp sleeve can be set, as well as a third class corresponding to the environment of the crimp connection.

[0200] Optionally, a fourth class can be set, corresponding to voids in the first class, i.e., voids in the region of the internal conductor. Optionally, at least one additional class can be set corresponding to specific sub-regions of internal components, such as multiple individual wires, or specific sub-regions of external components, such as burrs on the crimp sleeve.

[0201] The corresponding classes are typically planar regions of the cross-sectional image of the crimp connection. The vector contour determined from the raster view can also be formed as a planar region of the cross-sectional image, representing the determined classes.

[0202] The vector contour can also only relate to a part of the raster image, in particular at least one boundary region between adjacent classes. This boundary region or these boundary regions are typically very important for determining the quantitative and / or qualitative quality parameters of the crimp connection.

[0203] The output signal provided by the image processing device can in particular be configured to allow a graphical representation of planar and / or linear vector profiles.

[0204] The graphical representation can also be superimposed on the cross-sectional image of the received crimping connection, especially when displaying at least one boundary profile.

[0205] Furthermore, marking points that mark important structures of the cross-section of the crimping connection can be determined, such as reference points for the crimping width (average width) or the measurable crimping width (maximum crimping width), etc.

[0206] Figure 2 The structural schematic diagram of the image transformer network BTN configured as SegFormer is shown. Such an image transformer network BTN includes an encoder part E and a decoder part D. Generally, for the successful provision of a good enough semantic segmentation of the cross-sectional image, more importantly, especially the encoder part E and the multi-layer perceptron MLP associated with this image transformer network BTN. The decoder part D is used, among other things, to scale the result of the encoder part E, especially the number of pixels, to the original image size. However, the decoder part D can also have other or additional functions.

[0207] The cross-sectional image received by the receiving unit is fed to the encoder part E of the image transformer network. This can be processed by overlap patch embedding (OPE). In the overlap patch embedding OPE, multiple sub-regions of the image can be provided in several overlapping patches. For example, the overlapping area can be 50%.

[0208] Through overlapping, global and local context information is better captured, resulting in an improved context representation for prediction or segmentation. In addition, the image or sub-image is converted into a machine-readable form of the transformer block. In particular, the overlap patch embedding OPE helps to minimize artifacts that may occur due to the discrete nature of tile-based processing by supporting the continuity of features extending across adjacent tiles.

[0209] The overlapping patches of the cross-sectional image of the received crimping connection can be fed, for example, in vector form to the first transformer block T1. The structure of the BTN image transformer network shown here includes four transformer blocks T1, T2, T3, and T4, which are generally the same in structure and arranged in series. In addition, an overlap patch embedding OPE is provided before each transformer block.

[0210] Each of the four transformer blocks T1, T2, T3, and T4 includes a so-called attention mechanism, such as self-attention, attention matrices, or a mechanism for modeling the relationships between tokens or patches among them. These transformer blocks T1, T2, T3, and T4 use these attention mechanisms to process the provided sequence of patches to extract global and local features in the image and to model the relationships between the patches.

[0211] To this end, the four transformer blocks T1, T2, T3, and T4 can each include a module for efficient self-attention (ESA). This is a special form of self-attention that is characterized by using an alternative method to reduce the computational complexity of self-attention.

[0212] Approximate attention mechanisms and compressed representations are used to reduce the number of elements to be considered. The goal is to optimize the computation of self-attention without significantly degrading the model performance.

[0213] Thus, on the one hand, the efficient self-attention ESA promotes the scalability of SegFormer for large datasets and more complex tasks, which is why it can be easily adapted to further tasks. In addition, the computational resource requirements are lower than those of classical self-attention computations.

[0214] After the patches are computed by the efficient self-attention ESA, the generated patches are then fed into a mix-feed-forward network (MFFN), also known as the hybrid FFN. The hybrid-FFN MFFN performs a mixing operation based on the features of multiple individual patches. The goal of the hybrid FFN MFFN is to improve the representation within each patch by combining or transforming various features or information.

[0215] The operation of the hybrid FFN MFFN helps to emphasize or enhance specific features related to the semantic segmentation of the image by better highlighting or combining these features.

[0216] Subsequently, overlap patch merging (OPM) occurs, where information from the overlapping processed patches is merged or unified to achieve a more consistent and comprehensive representation of the image information. This allows for better integration of local and global context information.

[0217] In addition, these overlaps help reduce artifacts or discontinuities at the tile boundaries that might occur if tile - based processing does not account for the overlapping regions of the image. The overlapping tile merging OPM also helps reduce artifacts or discontinuities at the tile boundaries that might otherwise occur if tile - based processing does not consider the overlapping regions of the image.

[0218] Each transformer block T1, T2, T3, T4 has an output through which the data generated by the respective transformer block T1, T2, T3, T4 is provided. Then, each data is fed to another overlapping tile embedding OPE, and then this data has a modified overlapping tile partitioning, specifically multiple overlapping smaller tiles, and is fed in machine - readable form, such as as a vector, to a subsequent transformer block, such as T2. In this transformer block T2, the same structural steps as described above are performed. However, each transformer block T1, T2, T3, or T4 can also be of a different structure.

[0219] In this way, the data of the received cross - sectional image is continuously passed through four transformer blocks T1, T2, T3, and T4 in the form of an intermediate overlapping tile embedding (OPE).

[0220] (OPE).

[0221] Not only the final result formed by all four transformer blocks T1, T2, T3, T4, but also all intermediate results are fed to the multi - layer perceptron MLP. Thus, there are so - called shortcuts, also known as skip connections, which lead from the respective OPE step to the multi - layer perceptron MLP, and through which the respective data can be fed to the multi - layer perceptron MLP.

[0222] The multi - layer perceptron MLP supplements the transformation function of the transformer blocks T1, T2, T3, T4 by providing an additional level of feature modeling and processing that is specifically used to improve the segmentation accuracy. This allows the model to customize features according to the requirements of the segmentation task and transform them in a targeted way to achieve a more accurate segmentation.

[0223] Therefore, the multi - layer perceptron MLP is also part of the model, which can be adapted to the specific segmentation task here, i.e., the segmentation of the cross - sectional image of the crimp connector, by adding layers or by modifying existing layers (also known as fine - tuning) through subsequent training.

[0224] Through corresponding training, the multi - layer perceptron MLP of the image transformer network BTN can be further improved to enable at least one misclassification through the image transformer network. This means that in addition to image processing, further evaluation of the cross - sectional image of the crimp connector can be performed.

[0225] The multi - layer perceptron MLP after the transformer blocks T1, T2, T3, T4 in SegFormer is at least one additional layer applied to the outputs of the transformer blocks T1, T2, T3, T4 to improve the segmentation result and further optimize the model performance. This enables further improvement and adjustment of the representation of the cross - sectional image of the crimp connection obtained from the transformer blocks T1, T2, T3, T4.

[0226] The decoder D typically includes a pixel - class decoding layer that converts the extracted feature representation into pixel - class predictions. For example, this involves assigning multiple individual pixels to a class, such as pixels belonging to a first class, a second class, or a third class. Thus, this layer is very important for generating the desired raster view. This can be part of a multi - layer perceptron or configured separately.

[0227] In addition, on the decoder side D, there are typically upsampling operations and / or convolutional layers to increase the feature resolution and resize the predictions to the original input size of the image.

[0228] Typically, there is also a classification layer that predicts the probability or label of each pixel class in the image, thus creating a complete semantic segmentation map or segmentation mask.

[0229] The decoding layer, upsampling layer, and classification layer are jointly represented as a decoder module DM in Figure 2 .

[0230] For example, such a pre - trained SegFormer network can be found at https: / / huggingface.co / docs / transformers / model_doc / segformer.

[0231] Then, the model can be adjusted or customized for a specific application of semantic segmentation of the cross - sectional image of the crimp connection, possibly a crimp connection manufactured on a specific crimping device.

[0232] With the aid of such a model, a raster image of the semantic segmentation of the cross - sectional image of the crimp connection can be provided.

[0233] It should be understood that a person skilled in the art can also use other methods to perform semantic segmentation. In particular, a person skilled in the art can also use a deep neural convolutional network trained for the corresponding task.

[0234] Figure 3 Three exemplary images of the cross - section of the crimp connection C are shown, each represented as a photograph and a line drawing.

[0235] The first image B1 shows a representation of the received image, which is obtained by means of an image - processing device (e.g., according to Figure 1The object for image processing by an image processing device).

[0236] The second image B2 shows a raster image of semantic segmentation. The raster image includes a first class K1 corresponding to the internal conductor, a second class K2 corresponding to the crimp sleeve, and a third class K3 corresponding to the environment of the crimp connection. Each class K1, K2, K3 in the photograph is assigned a different color or shade. This makes the relative route of the boundary region between adjacent classes K1, K2, K3 clear.

[0237] The first class K1 has a first color F1, the second class K2 has a second color F2, and the third class K3 has a third color F3. Preferably, adjacent color regions have a high contrast so that they are visually easy to distinguish.

[0238] The second image B2 has been converted from a pixel-based raster view to a vector contour to increase the accuracy of potential measurements on the second image B2.

[0239] The third image B3 shows the received image, which is covered by a first boundary contour G1 and a second boundary contour G2. The first boundary contour G1 depicts the boundary route from the first class K1 to the second class K2. For example, this allows the distinction between the internal conductor region and the crimp sleeve. The second boundary contour G2 depicts the boundary route from the second class K2 to the third class K3, thus allowing, for example, the distinction between the crimp sleeve and the environment of the crimp connection.

[0240] The boundary contours G1 and G2 can be determined, for example, from the vector contour of the second image B2.

[0241] In addition, the third image B3 includes a plurality of marker points, in particular marker points M1, M2, M3, M4, and M5, which can be used to determine quantitative quality parameters of the crimp connection. The marker points M1, M2, M3, M4, and M5 can be determined conventionally based on the vector contour, for example, by logical operations or by a correspondingly trained deep neural network.

[0242] In the image B3, the exemplary marker points M1, M2, M3, M4, and M5, as well as additional marker points, are marked by white dots. However, different colors or other distinguishing features can be provided for each marker point M1, M2, M3, M4, and M5. In particular, all reference points can be provided with the marker points M1, M2, M3, M4, and M5, which are required to determine the quantitative quality parameters; see Figure 8 .

[0243] For example, based on the marker points M1 and M2, the (average) crimp width can be determined. For example, using the marker points M3, M4, and M5, the crimp height of the crimp connection can be determined.

[0244] Based on the corresponding marking points, a traceable and objective measurement of the quantitative quality parameters of the crimping connection C in the cross-section can be carried out.

[0245] Figure 4 An image evaluation device 200 is shown, which includes an image processing device 100 according to Figure 1 In the Figure 1 context, the statements regarding the image processing device 100 correspondingly apply to Figure 4 .

[0246] The output signal provided by the image processing device 100 is evaluated by the evaluation unit 201 of the image evaluation device 200.

[0247] For example, the evaluation unit 201 determines qualitative and / or quantitative quality parameters based on the marking points of the provided vector profile, such as the absence of defects in the crimping connection, crimping height, crimping width, measurable crimping width, support angle, support height, side end distance, crimping side end distance, burr height, burr width, bottom thickness, gap between wires, and / or crack. The evaluation unit 201 can also determine the marking points by itself without receiving these marking points from the image processing device.

[0248] In this process, the measured values of the corresponding quality parameters can first be determined from the vector profile of the quantitative quality parameters. Then, based on the corresponding scale, these measured values can be converted into real measurement units, such as degree and length measurements, for example, SI units.

[0249] In particular, the reference can be test bench-specific, thus taking into account the individual measurement setup for capturing the cross-sectional image of the crimping connection. Therefore, the evaluation of the output signal can include, for example, quantitative quality parameters that can be evaluated in subsequent comparisons.

[0250] However, in addition, the evaluation unit can also be configured such that it provides an evaluation output signal that includes a statement regarding the presence or absence of defects related to the analyzed crimping connection.

[0251] In particular, it can be determined whether the crimping connection is "normal", that is, without errors and within specific tolerances, or "abnormal", that is, exceeding specific tolerances, based on qualitative quality parameters and / or by comparing the determined quantitative quality parameters of the crimping connection with the corresponding target values.

[0252] In this configuration, the evaluation unit 201 is advantageously designed to be able to access the target values of specific quality parameters. In particular, an interface can be provided that is connected to a database ( Figure 4(not shown in the figure), where relevant manufacturing data, such as target values for specific quantitative quality parameters, are stored in the database and can be retrieved from the database. These values can then be used by the evaluation unit 201 for comparison with the corresponding actual values determined. In addition, the actual values determined by the image evaluation device 200 can be fed into the database and stored in the database for retrieval.

[0253] Figure 5 Another embodiment of the image evaluation device 200 is shown. It is characterized in that the image processing unit 102 and the evaluation unit 201 are combined into a combined image analysis unit 201'.

[0254] The combined image analysis unit 201' is designed such that it includes a trained deep neural network, in particular an image transformer network, which is designed to perform image processing and evaluation.

[0255] For this purpose, the deep neural network is trained accordingly, that is, by means of semantic segmentation of at least relevant image regions of the cross-sectional image, especially the entire cross-sectional image of the crimp connection, to generate a raster image. In addition, the deep neural network is designed to generate at least one vector contour based on the raster image and perform error classification based on the generated vector contour to determine specific qualitative and / or quantitative quality parameters of the crimp connection.

[0256] In addition, the trained deep neural network is designed to provide an evaluation output signal according to the performed error classification, in particular such that the evaluation output signal includes the performed error classification for determining specific qualitative and / or quantitative quality parameters of the crimp connection.

[0257] Such a trained deep neural network can have Figure 2 the structure shown, where a multi-layer perceptron is trained to perform the above tasks. For this purpose, additional layers can be added to the multi-layer perceptron and trained based on the misclassified training data.

[0258] By means of an exemplary configuration of the image evaluation device 200 according to Figure 4 or Figure 5 , a largely error-free automatic determination of quantitative and / or qualitative quality parameters can be implemented, thereby objectively measuring the characteristics of the crimp connection.

[0259] Figure 6 Possible results of the image evaluation device 200 are shown. In this regard, Figure 6 include image representations B1, B2, and B3 generated by an image processing device corresponding to Figure 3 .

[0260] In addition, Figure 6It includes a fourth image B4, which, for example, represents an actual value determined by an image transducer network for the crimp connection C to be inspected and / or a qualitative quality parameter such as "normal" or "abnormal".

[0261] For example, in particular, the fourth image B4 may have the following content.

[0262]

[0263]

[0264] The actual values in the above table are determined by a trained deep neural network, which provides a corresponding evaluation output signal that includes the corresponding actual value or description.

[0265] The evaluation may also include comparing whether the determined actual value lies within a specific tolerance range around the target value of the corresponding quality parameter of the crimp connection. The comparison evaluation can be performed independently of the trained deep neural network, for example, by a separate comparison unit, or it can also be performed by the trained deep neural network. This can be included in the image evaluation device.

[0266] Figure 7 A schematic diagram of a manufacturing release system 300 for a crimping device for manufacturing a crimp connection is shown.

[0267] The manufacturing release system 300 in this embodiment includes a cutting unit 301. The cable includes a crimp connection. In the cutting unit 301, a cross-section of the crimp connection can be generated by cutting and grinding the crimp connection perpendicular to the longitudinal direction of the cable.

[0268] In addition, the manufacturing release system 300 includes an image capture device 302, which is a microscope configured to capture a digital image of the cross-section of the crimp connection. The cut crimp connection is thus fed to the microscope 302 and captured imagewise by the microscope 302, particularly as precisely as possible and without image capture errors.

[0269] In addition, the manufacturing release system 300 includes an image evaluation device 200. For example, this can be configured according to Figure 4 or Figure 5 This includes a receiving unit 101 for receiving the cross-sectional image of the crimp connection.

[0270] In addition, this includes a trained deep neural network configured as an image transducer network, which is trained to perform image processing and evaluation. This can be performed by a combined image analysis unit 201'.

[0271] The combined image analysis unit 201’ is designed such that the required quantitative and qualitative quality parameters of the crimped connection can be determined by means of misclassification. The quantitative quality parameters can in particular include: crimp height, crimp width, measurable crimp width, support angle, support height, side end distance, crimp side end distance, burr height, burr width, bottom thickness, wire-to-wire gap, crack.

[0272] By means of the combined image analysis unit 201’, among other things, a quantitative determination of the quality parameters of the crimped connection is carried out, that is, a quantitative determination is carried out based on the measurement of distances and / or angles of the cross-sectional image of the crimped connection that has been captured and processed.

[0273] In the case of determining the quantitative quality parameters, any marking points used for the quantitative determination do not necessarily have to be visually displayable. However, the visual availability of the marking points used provides better traceability for the measured values determined.

[0274] The determined quantitative quality parameters are fed to a comparison unit 202, which is included in the image evaluation device 200 according to Figure 7 However, this is not necessary. The comparison unit 202 is connected to the manufacturing database 304 via a database interface 303, and the manufacturing database 304 is also included in the manufacturing release system 300. The manufacturing database 304 can be configured, for example, as a database, such as an ERP database, in particular an SAP database.

[0275] The manufacturing database 304 includes the relevant manufacturing data of production order A and thus includes the corresponding target values of the crimped connections according to production order A. Thus, the manufacturing database 304 can provide the target values of the required quantitative quality parameters and transmit them to the comparison unit 202.

[0276] The comparison unit 202 compares the determined quantitative quality parameters with the corresponding target values. Based on this comparison, it is clear whether the manufactured and analyzed crimped connections are defective. In addition, the determined quantitative quality parameters, that is, the corresponding actual values, are fed to the manufacturing database 304 and stored in the manufacturing database 304.

[0277] The comparison unit 202 is connected to a release unit 305. By means of the release unit 305, it can be determined whether the manufacture of the crimped connection should be released.

[0278] If this comparison shows that the crimped connection has a quantitative quality parameter within the target value tolerance, the manufacture of the crimped connection is released. If this comparison shows that the crimped connection has a quantitative quality parameter outside the target value tolerance, the manufacture of the crimped connection is not released. This can also apply to qualitative, i.e., non-measured, quality parameters, where if a specific quality parameter is qualitatively classified as "abnormal", the release unit 305 is not permitted to release.

[0279] Furthermore, the release unit 305 can be connected to a delivery system. If subsequent analysis of the cross-sectional image of the crimped connection points to a conclusion that the executed manufacturing order is defective, the release unit 305 can be designed to provide a signal to stop the delivery of the manufacturing order to the customer.

[0280] It should be understood that a person skilled in the art can also divide the functions of the aforementioned units of the manufacturing release system 300 differently, or, if required, set them up through a single technical unit.

[0281] Figure 8 A schematic cross-sectional view of a crimped connection is shown, showing reference points for measuring quantitative quality parameters.

[0282] Reference numeral 1 denotes a reference point of the crimped connection for measuring the crimp height; reference numeral 2 denotes a reference point of the crimped connection for measuring the crimp width. Reference numeral 3 denotes a reference point of the crimped connection for measuring the measurable crimp width. Reference numeral 4 denotes a reference point of the crimped connection for measuring the support angle. Reference numeral 5 denotes a reference point of the crimped connection for measuring the support height. Reference numeral 6 denotes a reference point of the crimped connection for measuring the side end distance. Reference numeral 7 denotes a reference point of the crimped connection for measuring the crimp side end distance. Reference numeral 8 denotes a reference point of the crimped connection for measuring the burr height. Reference numeral 9 denotes a reference point of the crimped connection for measuring the burr width. Reference numeral 10 denotes a reference point of the crimped connection for measuring the bottom thickness. Reference numeral 11 denotes a reference point of the crimped connection for measuring cracks. Although not shown, based on their professional knowledge, a person skilled in the art can easily determine the corresponding measured values of the voids in multiple directions based on multiple marked points of multiple voids. This also applies to the determination of cracks.

[0283] Since the above-described detailed equipment and methods are exemplary embodiments, a person skilled in the art can modify them to a great extent without departing from the scope of the present disclosure. In particular, the mechanical arrangement and dimensional relationships between the respective elements are merely exemplary.

[0284] List of reference numerals

[0285] 100 Image processing device

[0286] 101 Receiving Unit

[0287] 102 Processing Unit

[0288] 200 Image Evaluation Device

[0289] 201 Image Evaluation Unit

[0290] 201’ Combined Image Analysis Unit with a trained deep neural network for image processing and evaluation

[0291] 202 Comparison Unit

[0292] 300 Manufacturing Release System

[0293] 301 Cutting Unit

[0294] 302 Image Capture Device

[0295] 303 Database Interface

[0296] 304 Database

[0297] 305 Release Unit

[0298] K1 First Category: Internal Components of the Crimp Connection: Conductor Elements

[0299] K2 Second Category: External Components of the Crimp Connection: Crimp Sleeves

[0300] K3 Third Category: Environment of the Crimp Connection

[0301] B1 Received Cross-Sectional Image

[0302] B2 Three-Color Grating Image

[0303] B3 Overlay Image of the Received Image and the Boundary Contour

[0304] F1 Color 1

[0305] F2 Color 2

[0306] F3 Color 3

[0307] G1 Boundary Contour between the First Category and the Second Category G2 Boundary Contour between the Second Category and the Third Category C Cross-Section of the Crimp Connection

[0308] M1 Marking Point 1

[0309] M2 Marking Point 2

[0310] M3 Marking Point 3

[0311] M4 Marking Point 4

[0312] Marker point 5 of M5

[0313] Order number of A

[0314] Transformer block 1 of T1

[0315] Transformer block 2 of T2

[0316] Transformer block 3 of T3

[0317] Transformer block 4 of T4

[0318] Overlap Patch Embedding of OPE Multi-Layer Perceptron of MLP

[0319] Efficient Self Attention of ESA Mix-Feed Forward Network of MFFN Overlap Patch Merging of OPM Encoder of E

[0320] Decoder of D

[0321] Decoder module of DM

[0322] Image Transformer Network of BTN

[0323] 1 Crimping height

[0324] 2 Crimping width

[0325] 3 Measurable crimping width

[0326] 4 Support angle

[0327] 5 Support height

[0328] 6 Side end distance

[0329] 7 Crimping side end distance

[0330] 8 Burr height

[0331] 9 Burr width

[0332] 10 Bottom thickness

[0333] 11 Crack

Claims

1. An image processing device (100) for supporting qualitative and / or quantitative assessment of the quality of a crimp connection (C), comprising: a receiving unit (101) for receiving a cross-sectional representation of an image of the crimp connection (C), the cross-sectional representation being particularly configured as a digital microscope image in the visible spectral range; a processing unit (102) designed to generate a raster image of the received image based on the received image using a trained deep neural network, wherein the trained deep neural network assigns at least pixels of relevant image regions of the received image to pre-determined classes (K1, K2, K3), generate at least one vector contour (G1, G2) based on the generated raster image, generate and output an output signal based on the determined vector contour (G1, G2), and at least one qualitative and / or quantitative quality parameter (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11) assignable to the crimp connection (C) can be determined based on the output signal.

2. The image processing device according to claim 1, wherein the trained deep neural network comprises an image transformer network (BTN) and / or a convolutional network.

3. The image processing device according to claim 2, wherein the deep neural network is configured as an image transformer network (BTN), and the image transformer network (BTN) comprises a publicly available pre-trained image transformer network (BTN) and at least one network layer, and the at least one network layer is subsequently trained using application-specific training data for image processing.

4. The image processing device according to any one of the preceding claims, wherein the trained deep neural network is designed to generate a raster graphic based on the received image, the raster image having at least three classes (K1, K2, K3), wherein a first class (K1) corresponds to an internal component of the crimp connection (C), in particular a conductor element, a second class (K2) corresponds to an external component of the crimp connection (C), in particular a crimp sleeve, and a third class (K3) corresponds to the environment of the crimp sleeve, in particular the environment of the crimp connection (C).

5. The image processing device according to claim 4, wherein the trained deep neural network is designed to assign colors (F1, F2, F3) to at least the first class, the second class, and the third class (K1, K2, K3) respectively, wherein adjacent classes (K1, K2, K3) can be represented as color regions of different colors (F1, F2, F3), in particular color regions with distinct contrast, in the raster image and / or the vector contour, and this color assignment is included in the output signal.

6. The image processing device according to claim 4 or 5, wherein the trained deep neural network is designed to determine boundary contours (G1, G2) between the first class (K1) and the second class (K2) and / or between the second class (K2) and the third class (K3).

7. An image evaluation device (200) for qualitative and / or quantitative evaluation of the quality of a crimped connector (C), The image evaluation device (200) comprises an image processing device (100) according to any one of the preceding claims, Among them, Based on the generated vector contours, error classification can be performed, and the error classification is used to determine specific qualitative and / or quantitative quality parameters (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11) of the crimped connector (C), Wherein, an evaluation output signal can be generated and output according to the performed error classification, and the evaluation output signal especially includes the performed error classification for determining specific qualitative and / or quantitative quality parameters (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11) of the crimped connector (C).

8. The image evaluation device according to claim 7, Comprising a trained deep neural network, especially a trained deep image transformer network (BTN), and error classification can be performed by means of the trained deep neural network for determining specific qualitative and / or quantitative quality parameters (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11) of the crimped connector (C).

9. The image evaluation device according to claim 8, Among them, The image processing and the error classification for determining specific qualitative and / or quantitative quality parameters (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11) of the crimp joint (C) can be performed by a common deep neural network, especially an image transformer network (BTN).

10. The image evaluation device according to any one of claims 7 to 9, Wherein the evaluation output signal includes at least one of the following quality parameters (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11), especially at least one actual value from the following quality parameters (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11): defect-free crimped connector, crimp height (1), crimp width (2), measurable crimp width (3), support angle (4), support height (5), side end distance (6), crimp side end distance (7), burr height (8), burr width (8), bottom thickness (10), cavity between wires, crack.

11. The image evaluation device according to any one of claims 7 to 10, Wherein the evaluation output signal is designed to include an overlapping representation based on the received image and the vector contours (G1, G2), and the measurement points used for determining the corresponding quality parameters are included in the overlapping representation as specifically marked marker points (M1, M2, M3, M4, M5).

12. The image evaluation device according to any one of claims 7 to 11, The image evaluation device is designed to feed the evaluation output signal to a database (304), especially a manufacturing database.

13. A manufacturing release system (300) for a crimping device, the manufacturing release system (300) comprises: An image evaluation device (200) according to any one of claims 7 to 12; A data interface (303) that is connected to a database (304) in which production order-dependent target values for a crimp connection (C) are stored; A release unit (305) designed to provide release or rejection of a crimp connection (C) classified as having a manufacturing error after comparison of at least one qualitative and / or quantitative quality parameter (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11) with a corresponding target value.

14. The manufacturing release system according to claim 13, the manufacturing release system further comprising: A release unit (305) designed to provide release or rejection of the delivery of a classified crimp connection (C) based on the evaluation output signal.

15. The manufacturing release system according to any one of claims 13 or 14, the manufacturing release system further comprising: A database (304) through which the output signal can be stored, the output signal including at least one from the following parameters: defect-free crimp connection, crimp height (1), crimp width (2), measurable crimp width (3), support angle (4), support height (5), side end distance (6), crimp side end distance (7), burr height (8), burr width (9), bottom thickness (10), gap between conductors, crack.

Images