Screening method of object surface morphology based on artificial neural network

Abstract

A method for screening object surface morphology based on an artificial neural network is suitable for screening multiple objects. The method comprises: using multiple prediction models to perform surface morphology recognition on multiple object images to obtain a determined defect rate for each prediction model, wherein these object images correspond to the surface morphology of a portion of the objects; and, based on the determined defect rates of each prediction model, concatenating these prediction models into an artificial neural network system to screen the remaining objects. The method of screening object surface morphology based on an artificial neural network of the present invention concatenates multiple neural networks with different training conditions based on the determined defect rates of each neural network to provide an artificial neural network system that can accurately and quickly classify a large number of objects to be tested, while also maintaining a relatively high pass / fail rate.

Description

Technical field

[0001] The present invention relates to an artificial neural network training system, and more particularly to an object surface morphology screening method based on the artificial neural network. [Background Technology]

[0002] Various safety measures are composed of many small structural objects, such as seat belts. If these small structural objects are not strong enough, the protective effect of the safety measures may be questionable.

[0003] During the manufacturing process, these structural objects may develop tiny surface defects such as slots, cracks, bumps, and textures due to various reasons, such as collisions, process errors, and mold defects. These tiny defects are not easy to detect. One of the existing defect detection methods is to manually observe the structural object to be inspected with the naked eye or touch it with both hands to determine whether the structural object has defects such as pits, scratches, color differences, and defects. However, manual inspection of structural objects for defects is inefficient and is prone to misjudgment, which will make the yield of the structural object uncontrollable. [Summary of the invention]

[0004] In one embodiment, a method for screening object surface morphology based on an artificial neural network is suitable for screening multiple objects. The method includes: using multiple prediction models to perform surface morphology recognition on multiple object images to obtain a determined defect rate for each prediction model, where the object images correspond to the surface morphology of a portion of the objects; and, based on the determined defect rates of each prediction model, concatenating the prediction models into an artificial neural network system to screen the remaining objects.

[0005] In summary, according to the embodiment of the present invention, the method for screening object surface morphology based on an artificial neural network is implemented by connecting multiple neural networks with different training conditions based on the determined defect rate of each neural network to provide an artificial neural network system that can accurately and quickly classify a large number of objects to be tested while also achieving a better miss rate.

Brief Description of the Drawings

[0006] Figure 1 FIG. 4 is a flow chart of a method for screening object surface morphology based on an artificial neural network according to an embodiment of the present invention.

[0007] Figure 2 for Figure 1 FIG. 1 is a schematic diagram of an embodiment of step S02 in FIG.

[0008] Figure 3 FIG. 4 is a schematic diagram of an artificial neural network system according to an embodiment of the present invention.

[0009] Figure 4 FIG. 4 is a schematic diagram of an artificial neural network system according to another embodiment of the present invention.

[0010] Figure 5 for Figure 1 FIG. 1 is a schematic diagram of an embodiment of step S03 in FIG.

[0011] Figure 6 Flowchart of the training method of the sub-neural network system according to the first embodiment of the present invention.

[0012] Figure 7 Flowchart of the prediction method of the sub-neural network system according to the first embodiment of the present invention.

[0013] Figure 8 A schematic diagram of an example of an image region.

[0014] Figure 9 FIG. 4 is a flowchart of a training method for a sub-neural network system according to a second embodiment of the present invention.

[0015] Figure 10 Flowchart of the prediction method of the sub-neural network system according to the second embodiment of the present invention.

[0016] Figure 11 Flowchart of a training method for a sub-neural network system according to a third embodiment of the present invention.

[0017] Figure 12 Flowchart of a prediction method of a sub-neural network system according to the third embodiment of the present invention.

[0018] Figure 13 Flowchart of a training method for a sub-neural network system according to a fourth embodiment of the present invention.

[0019] Figure 14 Flowchart of a prediction method of a sub-neural network system according to a fourth embodiment of the present invention.

[0020] Figure 15 FIG. 1 is a schematic diagram of an example of an object image.

[0021] Figure 16 FIG. 1 is a schematic diagram of an image scanning system for object surface morphology according to a first embodiment of the present invention.

[0022] Figure 17 FIG. 4 is a functional diagram of an image scanning system for object surface morphology according to an embodiment of the present invention.

[0023] Figure 18 for Figure 16Schematic diagram of a first embodiment of the optical relative positions of an object, a light source assembly and a photosensitive element in FIG.

[0024] Figure 19 for Figure 16 Schematic diagram of a second embodiment of the optical relative positions of the object, light source assembly and photosensitive element in FIG.

[0025] Figure 20 FIG. 4 is a flow chart of an image scanning method for the surface morphology of an object according to a first embodiment of the present invention.

[0026] Figure 21 is a schematic diagram of an example of an object.

[0027] Figure 22 for Figure 21 A top view of the objects in the .

[0028] Figure 23 FIG. 4 is a flow chart of an image scanning method for the surface morphology of an object according to a second embodiment of the present invention.

[0029] Figure 24 FIG. 4 is a flow chart of an image scanning method for the surface morphology of an object according to a third embodiment of the present invention.

[0030] Figure 25 FIG. 1 is a schematic diagram of an example of detecting an image.

[0031] Figure 26 FIG. 4 is a partial flow chart of an image scanning method for the surface morphology of an object according to some embodiments of the present invention.

[0032] Figure 27 FIG. 4 is a partial flow chart of an image scanning method for the surface morphology of an object according to some other embodiments of the present invention.

[0033] Figure 28 for Figure 16 Schematic diagram of a third embodiment of the optical relative positions of an object, a light source assembly and a photosensitive element in FIG.

[0034] Figure 29 FIG. 1 is a schematic diagram of an embodiment of a surface morphology.

[0035] Figure 30 for Figure 16 Schematic diagram of a fourth embodiment of the optical relative positions of an object, a light source assembly and a photosensitive element in FIG.

[0036] Figure 31 for Figure 16 FIG. 5 is a schematic diagram of a fifth embodiment of the optical relative positions of an object, a light source assembly, and a photosensitive element in FIG.

[0037] Figure 32 FIG. 4 is a functional diagram of an image scanning system for object surface morphology according to another embodiment of the present invention.

[0038] Figure 33 for Figure 16 FIG. 6 is a schematic diagram of a sixth embodiment of the optical relative positions of an object, a light source assembly, and a photosensitive element in FIG.

[0039] Figure 34 FIG. 4 is a functional diagram of an image scanning system for object surface morphology according to another embodiment of the present invention.

[0040] Figure 35 for Figure 16 FIG. 7 is a schematic diagram of a seventh embodiment of the optical relative positions of an object, a light source assembly, and a photosensitive element in FIG.

[0041] Figure 36 FIG. 4 is a schematic diagram of another exemplary embodiment of detecting an image.

[0042] Figure 37 FIG. 4 is a schematic diagram of another exemplary embodiment of detecting an image.

[0043] Figure 38 FIG. 4 is a schematic diagram of an image scanning system for object surface morphology according to a second embodiment of the present invention.

[0044] Figure 39 FIG. 4 is a schematic diagram of an image scanning system for object surface morphology according to a third embodiment of the present invention.

[0045] Figure 40 FIG. 4 is a functional diagram of an image scanning system for object surface morphology according to yet another embodiment of the present invention.

[0046] Figure 41 FIG. 1 is a partial schematic diagram of an embodiment of an image scanning system for the surface morphology of an object.

[0047] Figures 42 to 45 They are Figure 41 Schematic diagram of the image obtained by lighting with four light source modules shown.

[0048] Figure 46 is a schematic diagram of an example of an initial image.

[0049] Figure 47 FIG. 4 is a schematic diagram of an image scanning system for object surface morphology according to a fourth embodiment of the present invention.

[0050] Figure 48 for Figure 47 Schematic diagram of a first embodiment of the optical relative positions of an object, a light source assembly and a photosensitive element in FIG.

[0051] Figure 49 for Figure 47 Schematic diagram of a second embodiment of the optical relative positions of the object, light source assembly and photosensitive element in FIG.

[0052] Figure 50 for Figure 47 Schematic diagram of a third embodiment of the optical relative positions of an object, a light source assembly and a photosensitive element in FIG.

[0053] Figure 51 FIG. 4 is a schematic diagram of another example of an object image.

[0054] Figure 52 FIG. 4 is a schematic diagram of another example of an object image. [Specific implementation method]

[0055] The method for screening the surface morphology of an object based on an artificial neural network is applicable to an artificial neural network system, wherein the artificial neural network system can be implemented on a processor.

[0056] In some embodiments, reference Figure 1 During the learning phase, the processor can execute deep learning on multiple sub-neural network systems (i.e., untrained artificial neural networks) using different training conditions to establish prediction models (i.e., trained artificial neural networks) for each of these sub-neural network systems to identify the surface morphology of an object, thereby obtaining trained sub-neural network systems (step S01). Here, these object images can be images of the surface of the same object at the same relative position. Furthermore, the artificial neural network system receives multiple object images using fixed imaging coordinate parameters. Furthermore, these batches of object images can be obtained by capturing images of the surfaces of multiple objects.

[0057] In some embodiments of step S01, at the beginning of the learning phase, the multiple sub-neural network systems may utilize the same neural network algorithm but with different parameters (e.g., different pre-processing, different number of layers, different number of neurons, or a combination thereof), or utilize different neural network algorithms with the same parameters, or utilize different neural network algorithms with different parameters, etc. For example, the processor may perform a deep learning process with different parameters on the same or different batches of object images to establish prediction models for the multiple sub-neural network systems. In another example, the processor may perform multiple deep learning processes on the same or different batches of object images to establish prediction models for the multiple sub-neural network systems.

[0058] In some embodiments, each sub-neural network system performs deep learning under different training conditions to establish its own prediction model. The training conditions may include, for example, different numbers of neural network layers, different neuron configurations, different input image preprocessing, different neural network algorithms, or any combination thereof. Image preprocessing may include feature enhancement, image cropping, data format conversion, image overlay, or any combination thereof. In some embodiments, the prediction model of each sub-neural network system may be implemented using the same or different neural network algorithms. The neural network algorithm may be, for example, a convolutional neural network (CNN) algorithm, but is not limited thereto.

[0059] Next, in the system creation phase, the processor connects these trained sub-neural network systems in series to form an artificial neural network system (step S02).

[0060] In an example of step S02, referring to Figure 2 After the prediction models of these sub-neural network systems are established, the processor feeds the same batch of object images into these sub-neural network systems, allowing the prediction models of these sub-neural network systems to classify these batches of object images individually to obtain the determined defect rates of these prediction models (step S02a). Then, the processor connects these prediction models in series according to the determined defect rates of each prediction model to form an artificial neural network system (step S02b). In other words, the multiple trained sub-neural network systems each perform surface morphology recognition on the same batch of object images to obtain the determined defect rates of each sub-neural network system. Then, the processor connects these sub-neural network systems in series according to the determined defect rates of these sub-neural network systems to obtain an artificial neural network system having multiple sub-neural network systems connected in series.

[0061] After the artificial neural network system is formed (step S02 or S02b), during the application phase, the processor can use the formed artificial neural network system to screen another batch of objects (step S03). In other words, the processor feeds the object images of the other batch of objects into the artificial neural network system, which then performs classification predictions based on the fed object images. For example, the processor pre-trains multiple sub-neural network systems. Furthermore, the processor receives object images of multiple objects. When the artificial neural network system is created (i.e., during the system creation phase), the processor feeds the object images of a portion of objects into the multiple trained sub-neural network systems, allowing each sub-neural network system to screen these objects and obtain its determined defect rate. At this point, these sub-neural network systems each perform classification predictions based on the object images of this portion of objects and obtain their respective determined defect rates based on the classifications. Next, the processor connects these sub-neural network systems in series into a single sub-neural network system based on their determined defect rates. Then, when the artificial neural network system is applied (ie, the application stage), the processor feeds the object images of the remaining objects into the artificial neural network system to screen the remaining objects.

[0062] Therefore, the artificial neural network system connects multiple neural networks with different training conditions based on the determination defect rate of each neural network to provide an artificial neural network system that can accurately and quickly classify a large number of objects to be tested while taking into account a better miss rate.

[0063] Reference Figure 3 The artificial neural network system 30 may include an input unit 31, multiple sub-neural network systems 33, and an output unit 35. The sub-neural network systems 33 are sequentially connected between the input unit 31 and the output unit 35, and each sub-neural network system 30 is connected to the sub-neural network system 33 of the next level in series based on a portion of its output. Each sub-neural network system 33 has a prediction model.

[0064] In some embodiments, the output of each sub-neural network system 33 can be divided into a normal group and an abnormal group, and the normal group of each sub-neural network system 33 is coupled to the input of the next sub-neural network system 33. Figure 1 and Figure 3In the application stage, the object image IM fed into the artificial neural network system 30 is sequentially screened by each sub-neural network system 33 (step S03). For example, in the application stage, when one or more object images IM are fed into the artificial neural network system 30, the first-level sub-neural network system 33 executes a prediction model on each object image IM to classify it into a first-level normal group or a first-level abnormal group. When classified as the first-level normal group, the object image IM output to the first-level normal group by the first-level sub-neural network system 33 is continuously fed into the second-level sub-neural network system 33, so that the second-level sub-neural network system 33 continuously executes a prediction model on the object image IM to classify it into a second-level normal group or a second-level abnormal group. Conversely, when classified as the first-level abnormal group, the object image IM output to the first-level normal group by the first-level sub-neural network system 33 is not fed into the second-level sub-neural network system 33. This process is deduced in this way until the last-stage sub-neural network system 33 executes the prediction model on the object image IM fed from the previous stage (ie, the object image IM classified as the normal group of the previous-stage sub-neural network system 33).

[0065] In some embodiments, the output unit 35 receives the abnormal groups output by all sub-neural network systems 33 and outputs an abnormal result accordingly, and the output unit 35 also receives the normal group output by the last-level sub-neural network system 33 and outputs a normal result accordingly.

[0066] For the sake of convenience, two sub-neural network systems 33 are used as an example, but this number is not limited to the present invention. Figure 4 The two sub-neural network systems 33 are respectively called the first sub-neural network system 33a and the second sub-neural network system 33b.

[0067] The input of the first neural network subsystem 33a is coupled to the input unit 31. A portion of the output of the first neural network subsystem 33a is coupled to the input of the second neural network subsystem 33b, and another portion of the output of the first neural network subsystem 33a is coupled to the output unit 35.

[0068] Here, the first neural network subsystem 33a has a first prediction model. The second neural network subsystem 33b has a second prediction model. In some embodiments, the first prediction model can be implemented using a CNN algorithm. The second prediction model can also be implemented using a CNN algorithm. However, this is not a limitation of the present invention.

[0069] Here, refer to Figure 4 and Figure 5The input unit 31 receives one or more object images IM (step S03a) and feeds the received object images IM into the first sub-neural network system 33a. Next, the first prediction model of the first sub-neural network system 33a performs surface morphology recognition on each object image IM to classify it into one of a first normal group G12 and a first abnormal group G11 (step S03b). In other words, after the first prediction model recognizes the surface morphology of the object image IM, it classifies the object image IM into the first normal group G12 or the first abnormal group G11 based on the recognition result.

[0070] The object image IM classified as the first normal group G12 is then fed into the second neural network sub-system 33b, where the second prediction model of the second neural network sub-system 33b performs surface morphology recognition to classify the object image IM into one of a second normal group G22 and a second abnormal group G21 (step S03c). In other words, the second prediction model recognizes the surface morphology of the object image IM belonging to the first normal group G12 and then classifies the object image IM into the second normal group G22 or the second abnormal group G21 based on the recognition result.

[0071] Finally, the output unit 35 receives the first abnormal group G11 output by the first prediction model, the second abnormal group G21 output by the second prediction model, and the second normal group G22 output by the second prediction model, and outputs an abnormal result and a normal result. The abnormal result includes object images IM classified as belonging to the first abnormal group G11 and object images IM classified as belonging to the second abnormal group G21. The normal result includes object images IM classified as belonging to the second normal group G22.

[0072] In some embodiments, the number of neural networks connected in series in the artificial neural network system 30 can be designed to be 2 neural networks, 3 neural networks, 4 neural networks or more connected in series according to actual needs.

[0073] In some embodiments, the processor can cascade multiple sub-neural network systems 33 into an artificial neural network system 30 based on the judgment defect rates of their prediction models, from high to low. For example, sub-neural network systems 33 with higher judgment defect rates are arranged first, while sub-neural network systems 33 with lower judgment defect rates are arranged last. In other words, the judgment defect rates of the multiple cascaded sub-neural network systems 33 decrease in sequence. Therefore, during the application phase, the artificial neural network system 30 prioritizes objects with higher judgment defect rates based on the prediction models. Based on this, the artificial neural network system 30 can quickly classify and predict a large number of objects to be tested while also taking into account a high miss rate.

[0074] Here, when the surface of an object has any surface morphology, the corresponding image position of the object image of the object also includes an image of the surface morphology. For example, when the surface of an object has a sand hole, the sand hole will also be imaged at the corresponding image position of the object image of the object. When the surface of an object has a bump, the bump will also be imaged at the corresponding image position of the object image of the object. In some embodiments, the surface morphology may be a surface structure such as a slot, crack, bump, sand hole, air hole, bump, scratch, edge, or texture. Each surface structure is a three-dimensional microstructure. Here, the three-dimensional microstructure is sub-micron to micron (μm) in size. That is, the longest side or longest diameter of the three-dimensional microstructure is between sub-micron and micron. Sub-micron refers to less than 1 μm, for example, 0.1 μm to 1 μm. For example, the three-dimensional structure can be a microstructure ranging from 300 nm to 6 μm.

[0075] In some embodiments, at least one of the sub-neural network systems 33 may perform image pre-processing for image cropping.

[0076] Reference Figure 6 During the learning phase, the sub-neural network system 33 receives a plurality of object images IM (step S11). Here, these object images are all images of the surface of the same object at the same relative position. Next, the sub-neural network system 33 divides each object image IM into a plurality of image regions (step S12) and designates at least one region of interest (ROI) in the plurality of image regions of each object image IM (step S13). In other words, after an object image IM is cut into a plurality of image regions, the sub-neural network system 33 can designate corresponding image regions in the plurality of image regions as ROIs according to a specified setting. Then, the sub-neural network system 33 performs deep learning (training) using the designated ROIs to establish a prediction model for recognizing the surface morphology of the object (step S14). In some embodiments, the sub-neural network system 33 can perform division, designation, and training for each image one by one. In other embodiments, the sub-neural network system 33 can first divide and designate each object image and then train all designated ROIs at once.

[0077] In the prediction phase (i.e., the system creation phase or the application phase), the sub-neural network system 33 performs classification prediction using substantially the same steps as in the learning phase. Figure 7, the sub-neural network system 33 receives one or more object images IM (step S21). Here, the imaging target and position of each object image IM will be the same as the imaging target and imaging position of the object image IM used in the learning phase (such as the same relative position of the same object). Next, the sub-neural network system 33 divides each object image IM into multiple image areas (step S22), and designates at least one region of interest among the multiple image areas of each object image IM (step S23). In other words, after an object image IM is cut into multiple image areas, the sub-neural network system 33 can designate the image areas of corresponding order among the multiple image areas as regions of interest according to the specified settings. Then, the sub-neural network system 33 executes the prediction model with the designated region of interest to identify the surface morphology of the object (step S24).

[0078] Based on this, the sub-neural network system 33 can flexibly import the detection results of a specific area (specified region of interest). In some embodiments, the sub-neural network system 33 can also obtain a lower over-discharge rate, such as an over-discharge rate close to zero.

[0079] In some embodiments, the number of image regions into which each object image IM is divided is any integer greater than 2. Preferably, the image size of each image region is less than or equal to 768*768 pixels, such as 400*400 pixels, 416*416 pixels, 608*608 pixels, etc. Furthermore, the image regions all have the same image size. In some embodiments, each image region is preferably square. For example, if the image size of the object image IM is 3000*4000 pixels, the image size of the cropped image region can be 200*200 pixels.

[0080] In some embodiments of step S12 (or step S22), the sub-neural network system 33 may first enlarge the object image IM according to a preset cropping size, so that the size of the object image IM is an integer multiple of the size of the image region. The sub-neural network system 33 then crops the enlarged object image IM into a plurality of image regions according to the preset cropping size. Here, the image size of each image region is the same, i.e., the preset cropping size.

[0081] For example, refer to Figure 8 The neural network subsystem 33 divides each received object image IM into 70 image regions A01-A70 using the same cropping size. The neural network subsystem 33 then designates image regions A01-A10 as regions of interest (ROIs) based on a preset designation (assuming the designation is 1-10). The neural network subsystem 33 then performs deep learning or executes a prediction model based on image regions A01-A10 (i.e., the ROIs).

[0082] In some embodiments, the regions of interest may be imaged regions with sand holes of varying depths, imaged regions without sand holes but with bumps or scratches, imaged regions with varying surface roughness, imaged regions without surface defects, or imaged regions with defects of varying aspect ratios. Sub-neural network system 33 performs deep learning or executes a prediction model based on the aforementioned regions of interest with varying surface morphologies. During the learning phase, sub-neural network system 33 may categorize regions of interest with varying surface morphologies to generate different pre-defined surface morphology categories.

[0083] For example, sub-neural network system 33 can use regions of interest to identify that region of interest A01 is imaged with sand holes and bumps, region of interest A02 is not imaged and has defects, and region of interest A33 is imaged with only sand holes, and the imaged surface roughness is less than the surface roughness of region of interest A35. In the prediction stage, taking the five preset surface morphology categories of sand holes or pores, scratches or bumps, high roughness, low roughness, and no surface defects as an example, sub-neural network system 33 can classify region of interest A01 into the preset categories of sand holes or pores and scratches or bumps, region of interest A02 into the preset category of no surface defects, region of interest A33 into the preset categories of sand holes or pores and low roughness, and region of interest A35 into the preset category of high roughness.

[0084] In one embodiment of step S13 (or step S23), for each object image IM, the sub-neural network system 33 specifies the region of interest by changing the weight of each image region. Figure 8 After the object image IM is cropped into a plurality of image regions A01-A70, the weights of the plurality of image regions A01-A70 are initially preset to 1. In one embodiment, assuming that the designated settings are 1-5, 33-38, and 66-70, the sub-neural network system 33 increases the weights of the image regions A1-A5, A33-A38, and A66-A70 to 2 according to the preset designated settings, thereby designating the image regions A1-A5, A33-A38, and A66-A70 as regions of interest. In one example, when the weights of the regions of interest are increased, the weights of the other image regions A6-A32 and A39-A65 may be maintained at 1. In another example, when the weights of the regions of interest are increased, the sub-neural network system 33 may simultaneously decrease the weights of the other image regions A6-A32 and A39-A65 to 0.

[0085] In another embodiment, assuming that the designated settings are 1-5, 33-38, and 66-70, the artificial neural network system 30 reduces the weights of the image areas A6-A32, A39-A65 other than the image areas A1-A5, A33-A38, A66-A70 to 0 or 0.5 according to the preset designated settings, while the weights of the image areas A1-A5, A33-A38, A66-A70 remain at 1, thereby designating the image areas A1-A5, A33-A38, A66-A70 as the regions of interest.

[0086] In one embodiment, the sub-neural network system 33 may include a pre-processing unit and a deep learning unit. The input of the pre-processing unit is coupled to the previous stage of the sub-neural network system 33 (the previous sub-neural network system 33 or the input unit 31), and the output of the pre-processing unit is coupled to the input of the deep learning unit. The output of the deep learning unit is coupled to the next stage of the sub-neural network system 33 (the next sub-neural network system 33 or the output unit 35). Here, the pre-processing unit is used to perform the aforementioned steps S11 to S13 or steps S21 to S23, and the deep learning unit is used to perform the aforementioned steps S14 or S24. In other words, the architecture of the deep learning unit after performing deep learning is a prediction model. In another embodiment, the deep learning unit may include an input layer and multiple hidden layers. The input layer is coupled between the previous stage (the previous sub-neural network system 33 or the input unit 31) and each hidden layer. Each hidden layer is coupled between the input layer and the next stage (the next sub-neural network system 33 or the output unit 35). Here, the aforementioned steps S11 to S13 or steps S21 to S23 may be performed by the input layer instead.

[0087] In some embodiments, at least one of the sub-neural network systems 33 may perform image preprocessing to convert data formats.

[0088] Reference Figure 9 During the learning phase, the neural network sub-system 33 receives a plurality of object images IM (step S31). Next, the neural network sub-system 33 converts the object images IM into matrices based on their color patterns (step S32). This involves converting the object image data format into a format supported by the input channel of the artificial neural network (e.g., an image matrix). The neural network sub-system 33 then performs deep learning on the matrices to establish a predictive model for identifying the surface morphology of the object (step S33).

[0089] Here, the received object images IM are all images of the surface of the same object at the same relative position. The received object images IM have multiple color modes, and each object image IM has one of these color modes. In some embodiments, these color modes may include multiple different spectra. For example, during the learning phase, the processor can feed a large number of object images IM to the sub-neural network system 33. The fed object images IM include surface images of different spectra at the same relative position of multiple objects 2 of the same object 2 (i.e., object images IM).

[0090] Here, the artificial neural network in the sub-neural network system 33 has multiple image matrix input channels for inputting corresponding matrices, and these image matrix input channels each represent multiple imaging conditions (e.g., multiple color modes). In other words, the sub-neural network system 33 converts object images IM of various color modes into information such as length, width, pixel type, pixel depth, and number of channels within the matrix, where the number of channels represents the imaging conditions of the corresponding object image. Furthermore, the converted matrix is imported into the corresponding image matrix input channel based on the color mode of the object image to facilitate deep learning. In some embodiments, these image matrix input channels each represent a plurality of different spectra.

[0091] In some embodiments, the plurality of spectra may range from 380 nm to 3000 nm. For example, the plurality of different spectra may be any of visible light, such as white light, violet light, blue light, green light, yellow light, orange light, and red light. In one embodiment, the wavelength of white light may be between 380 nm and 780 nm, the wavelength of violet light may be between 380 nm and 450 nm, the wavelength of blue light may be between 450 nm and 495 nm, the wavelength of green light may be between 495 nm and 570 nm, the wavelength of yellow light may be between 570 nm and 590 nm, the wavelength of orange light may be between 590 nm and 620 nm, and the wavelength of red light may be between 620 nm and 780 nm. In another exemplary embodiment, the spectrum may be far-infrared light, having a wavelength between 800 nm and 3000 nm.

[0092] In some embodiments, these color modes may further include a grayscale mode, in which the object image IM is first converted into a grayscale image and then converted into a matrix having a number of channels representing grayscales.

[0093] In the prediction phase, the sub-neural network system 33 performs classification prediction using substantially the same steps as in the learning phase. Figure 10The sub-neural network system 33 receives one or more object images IM (step S41). Here, each object image IM is an image of the surface of the same object at the same relative position and has a specific color pattern. Next, the sub-neural network system 33 converts the object image IM into a matrix based on the color pattern of the object image IM (step S42). The sub-neural network system 33 then executes a prediction model using the matrix to identify the surface shape of the object (step S43).

[0094] In some embodiments, the sub-neural network system 33 may first normalize the object image IM to reduce the asymmetry between learning data and improve learning efficiency. Then, the sub-neural network system 33 converts the normalized object image IM into a matrix.

[0095] Based on this, the sub-neural network system 33 performs deep learning using a matrix having a number of channels representing different color modes, thereby establishing a prediction model that can identify information such as the structural type and surface texture (i.e., surface type) of the surface 21 of the object 2. In other words, by controlling the light emission spectrum or the light reception spectrum to provide object images with different imaging effects of the same object, the sub-neural network system 33 can improve its ability to distinguish various target surface types of the object. In some embodiments, this sub-neural network system 33 can integrate multi-spectral surface texture images to enhance the recognition of the target surface type of the object, thereby obtaining the surface roughness and fine texture type of the object.

[0096] In one embodiment, the sub-neural network system 33 may include a pre-processing unit and a deep learning unit. The input of the pre-processing unit is coupled to the previous stage of the sub-neural network system 33 (the previous sub-neural network system 33 or the input unit 31), and the output of the pre-processing unit is coupled to the input of the deep learning unit. The output of the deep learning unit is coupled to the next stage of the sub-neural network system 33 (the next sub-neural network system 33 or the output unit 35). Here, the pre-processing unit is used to perform the aforementioned steps S31 to S32 or steps S41 to S42, and the deep learning unit is used to perform the aforementioned steps S33 or S43. In other words, the architecture of the deep learning unit after performing deep learning is a prediction model. In another embodiment, the deep learning unit may include an input layer and multiple hidden layers. The input layer is coupled between the previous stage (the previous sub-neural network system 33 or the input unit 31) and each hidden layer. Each hidden layer is coupled between the input layer and the next stage (the next sub-neural network system 33 or the output unit 35). Here, the aforementioned steps S31 to S32 or steps S41 to S42 may be performed by the input layer instead.

[0097] In some embodiments, at least one of the sub-neural network systems 33 may perform image pre-processing for image superposition.

[0098] In one embodiment, referring to Figure 11 During the learning phase, the sub-neural network system 33 receives multiple object images IM of multiple objects (step S51). These object images IM are all images of the surface of the same object at the same relative position. The multiple object images IM of the same object are obtained by capturing the image of the object based on light from different lighting directions. In one example, the captured image of the same object may have the same spectrum or may have multiple different spectra. Next, the sub-neural network system 33 superimposes the multiple object images IM of each object into a superimposed object image (hereinafter referred to as the initial image) (step S52). Then, the sub-neural network system 33 performs a deep learning on the initial images of each object to establish a prediction model for identifying the surface morphology of the object (step S54). For example, the received object images IM include multiple object images IM of a first object and multiple object images IM of a second object. The sub-neural network system 33 superimposes multiple object images IM of the first object into an initial image of the first object and superimposes multiple object images IM of the second object into an initial image of the second object, and then performs deep learning with the initial image of the first object and the initial image of the second object.

[0099] In the prediction phase, the sub-neural network system 33 performs classification prediction using substantially the same steps as in the learning phase. Figure 12 , the sub-neural network system 33 receives multiple object images IM of an object (step S61). Here, the multiple object images IM of the object are all images of the surface of the object at the same position. Moreover, the multiple object images IM of the object are images of the object based on light from different lighting directions. Next, the sub-neural network system 33 superimposes the multiple object images IM of the object into an initial image (step S62). Then, the sub-neural network system 33 executes a prediction model using the initial image to identify the surface shape of the object (step S64).

[0100] Based on this, the sub-neural network system 33 can be trained by taking images from multiple angles (i.e., different lighting directions) in conjunction with multi-dimensional superposition pre-processing to improve the recognition of the three-dimensional structural features of the object without increasing the calculation time. In other words, by controlling the various incident angles of the imaging light source to provide object images with different imaging effects for the same object, the sub-neural network system 33 can improve the spatial three-dimensional distinction of the various surface forms of the object. Moreover, by integrating the object images under different lighting directions and superimposing the object images in multiple dimensions, the sub-neural network system 33 can improve its recognition of the surface form of the object, thereby obtaining the best analysis of the surface form of the object.

[0101] In an example of step S52 (or step S62 ), superimposing refers to adding the brightness values of each pixel in the object image IM.

[0102] In another embodiment, referring to Figure 13 and Figure 14 After step S52 or S62, the sub-neural network system 33 may first convert the initial image of each object into a matrix (step S53 or S63), that is, convert the data format of the initial image of each object into a format supported by the input channel of the artificial neural network (such as an image matrix). Then, the sub-neural network system 33 executes a deep learning or prediction model with the matrix of each object (step S54' or S64'). In other words, the sub-neural network system 33 converts the initial image of each object into information such as length, width, pixel type, pixel depth, and number of channels within the matrix, where the number of channels represents the color mode of the corresponding initial image. In addition, the converted matrix will be imported into the corresponding image matrix input channel according to the color mode of the initial image to facilitate the next step of processing.

[0103] In an example of step S52 (or step S62), the sub-neural network system 33 first normalizes the received object image IM and then superimposes the normalized object images IM of the same object into the initial image. This can reduce the asymmetry between the learning data and improve learning efficiency.

[0104] In one example of step S51 (or step S61), the object images IM of the same object may have the same spectrum. In another example of step S51 (or step S61), the object images IM of the same object may have multiple different spectra. In other words, the multiple object images IM of the same object include images of the object captured using light of one spectrum at different lighting directions, and images of the object captured using light of another spectrum at different lighting directions. Furthermore, these two spectra are different from each other.

[0105] In one embodiment, the sub-neural network system 33 may include a pre-processing unit and a deep learning unit. The input of the pre-processing unit is coupled to the previous stage of the sub-neural network system 33 (the previous sub-neural network system 33 or the input unit 31), and the output of the pre-processing unit is coupled to the input of the deep learning unit. The output of the deep learning unit is coupled to the next stage of the sub-neural network system 33 (the next sub-neural network system 33 or the output unit 35). Here, the pre-processing unit is used to perform the aforementioned steps S51-S53 or steps S61-S63, while the deep learning unit is used to perform the aforementioned steps S54, S54', S64, or S64'. In other words, the architecture of the deep learning unit after performing deep learning is a prediction model. In another embodiment, the deep learning unit may include an input layer and multiple hidden layers. The input layer is coupled between the previous stage (the previous sub-neural network system 33 or the input unit 31) and each hidden layer. Each hidden layer is coupled between the input layer and the next stage (the next sub-neural network system 33 or the output unit 35). Here, the aforementioned steps S51 to S53 or steps S61 to S63 may be performed by the input layer instead.

[0106] In some embodiments, each object image IM is formed by stitching together a plurality of detection images MB (eg Figure 15 In one example, the image size of the aforementioned region of interest is smaller than the image size of the detection image (the original image size).

[0107] In some embodiments, each detection image MB may be generated by performing an image scanning on the object 2 using an image scanning system specific to the surface morphology of the object.

[0108] Reference Figure 16 An image scanning system for object surface morphology is adapted to scan an object 2 to obtain at least one detection image MB of the object 2. Here, the object 2 has a surface 21, and along an extension direction D1 of the surface 21 of the object 2, the surface 21 of the object 2 is divided into a plurality of surface blocks 21A-21C. In some embodiments, the surface 21 of the object 2 is divided into nine surface blocks, three of which are exemplarily shown in the figure. However, this is not limiting, and the surface 21 of the object 2 may also be divided into any other number of surface blocks, such as 3, 5, 11, 15, 20, or any other number, depending on actual needs.

[0109] Reference Figures 16 to 19 , Figure 18 and Figure 19 They are Figure 16Schematic diagrams of two embodiments illustrating the optical relative positions of an object 2, a light source assembly 12, and a photosensitive element 13. An image scanning system for object surface topography includes a drive assembly 11, a light source assembly 12, and a photosensitive element 13. The light source assembly 12 and the photosensitive element 13 face a detection position 14 on the drive assembly 11 at different angles.

[0110] The image scanning system can execute an image acquisition process. Figures 16 to 20 During the image capture process, the driver assembly 11 carries the object 2 to be inspected and sequentially moves one of the plurality of surface segments 21A-21C to the aforementioned inspection position 14 (step S110). Furthermore, the light source assembly 12 emits a light beam L1 toward the inspection position 14 (step S120) to illuminate the inspection position 14 from the front or side. In this manner, the surface segments 21A-21C are sequentially positioned at the inspection position 14 and, while at the inspection position 14, are illuminated by the light beam L1 from the side or oblique direction.

[0111] In some embodiments, when each of the surface blocks 21A-21C is located at the detection position 14, the photosensitive element 13 receives diffuse light generated by the surface block currently located at the detection position 14, and captures a detection image of the surface block currently located at the detection position 14 based on the received diffuse light (step S130).

[0112] In some embodiments, the image scanning system may further include a processor 15. The processor 15 is coupled to the light source assembly 12, the photosensitive element 13, and the driving motor 112, and is used to control the operation of each assembly (eg, the light source assembly 12, the photosensitive element 13, and the driving motor 112).

[0113] In some embodiments, after the photosensitive element 13 captures the detection images MB of all the surface blocks 21A- 21C of the object 2 , the processor 15 can perform a stitching process based on the detection images MB to obtain the object image IM of the object 2 (step S140 ).

[0114] For example, in the image capturing process, the driving component 11 first moves the surface block 21A to the detection position 14, and when the detection light L1 provided by the light source component 12 illuminates the surface block 21A, the photosensitive element 13 captures the detection image Ma of the surface block 21A, such as Figure 15 Then, the driving component 11 moves the object 2 to move the surface block 21B to the detection position 14, and the detection light L1 provided by the light source component 12 illuminates the surface block 21B, and the photosensitive element 13 captures the detection image Mb of the surface block 21B, as shown. Figure 15Next, the driving component 11 moves the object 2 to move the surface block 21C to the detection position 14, and the detection light L1 provided by the light source component 12 illuminates the surface block 21B, and the photosensitive element 13 captures the detection image Mc of the surface block 21C, as shown. Figure 15 As shown. And so on, until all the detection images MB of the surface blocks are captured. Then, the processor 15 stitches these detection images MB into an object image IM, as shown Figure 15 shown.

[0115] In some embodiments, specifically, during the image capture process, object 2 is supported on drive assembly 11, and one of surface areas 21A-21C of object 2 is substantially located at detection position 14. Prior to each image capture, the image scanning system performs alignment (i.e., fine-tunes the position of object 2) to align the surface area with the viewing angle of photosensitive element 13.

[0116] In some embodiments, reference Figure 21 and Figure 22 The object 2 includes a main body 201, a plurality of first alignment structures 202, and a plurality of second alignment structures 203. The first alignment structure 202 is located at one end of the main body 201, and the second alignment structure 203 is located at the other end of the main body 201. In some embodiments, the first alignment structure 202 may be a structure such as a column, a protrusion, a slot, etc. The second alignment structure 203 may be a structure such as a column, a protrusion, a slot, etc. In some embodiments, the second alignment structures 203 are arranged at intervals along the extension direction of the surface 21 of the main body 201 (hereinafter referred to as the first direction D1), and the spacing distance between any two adjacent second alignment structures 203 is greater than or equal to the viewing angle of the photosensitive element 13. In some embodiments, these second alignment structures 203 correspond to surface blocks 21A-21C of the object 2, respectively. Each second alignment structure 203 is aligned with the middle of the side of its corresponding surface block along the first direction D1.

[0117] The following example uses a first alignment structure 202 as a column (hereinafter referred to as an alignment column) and a second alignment structure 203 as a slot (hereinafter referred to as an alignment slot). In some embodiments, the extension direction of each alignment column is substantially the same as the extension direction of the body 201, and one end of each alignment column is coupled to one end of the body 201. The alignment slots are located at the other end of the body 201, circumventing the body 201 with the long axis of the body 201 as the rotation axis, and are spaced apart on the surface of the other end of the body 201.

[0118] In some embodiments, the first alignment structures 202 are spaced apart on the body 201. In this example, three first alignment structures 202 are used as an example, but this number is not a limitation of the present invention. When looking down at the side of the body 201, as the body 201 rotates about its long axis, the first alignment structures 202 will present different relative positions. For example, the first alignment structures 202 are spaced apart and do not overlap each other (e.g., Figure 22 ), or any two first alignment structures 202 overlap but the remaining first alignment structure 202 does not overlap, etc.

[0119] In some embodiments, please refer to Figures 16 to 19 and Figures 21 to 24 In the image capture process, under the illumination of the light source assembly 12 , the processor 15 controls the photosensitive element 13 to capture a test image of the object 2 (step S211 ). Here, the test image includes an image block showing the second alignment structure 203 currently facing the photosensitive element 13 .

[0120] The processor 15 detects the position of the image block presenting the second alignment structure 203 in the test image (step S212 ) to determine whether the surface block currently located at the detection position 14 is aligned with the viewing angle of the photosensitive element 13 .

[0121] If the image block is not positioned in the center of the test image, processor 15 controls drive assembly 11 to fine-tune the position of object 2 in first direction D1 (step S213), and then returns to step S211. Steps S211 to S213 are repeatedly executed until processor 15 detects that the image block is positioned in the center of the test image.

[0122] When the image block is positioned in the middle of the test image, the processor 15 drives the photosensitive element 13 to capture an image. At this time, the photosensitive element 13 captures a detection image of the surface block of the object 2 under the illumination of the light source assembly 12 (step S214 ).

[0123] Next, the processor 15 controls the drive assembly 11 to move the next surface area of the object 2 in a first direction to the detection position 14, so that the next second alignment structure 203 faces the photosensitive element 13 (step S215). The process then returns to step S211. Steps S211 to S215 are repeatedly executed until detection images of all surface areas of the object 2 are captured. In some embodiments, the amplitude by which the drive assembly 11 fine-tunes the object 2 is smaller than the amplitude used to move the next surface area of the object 2.

[0124] Based on this, the image scanning system can utilize image analysis to test the specific structure of the object in the image to determine whether the object is aligned, thereby obtaining an image aligned with the viewing angle of the sensor element 13 .

[0125] For example, assume that object 2 has three surface areas, and the photosensitive element 13 faces surface area 21A of object 2 when the image capture process begins. At this point, under the illumination of light source assembly 12, photosensitive element 13 first captures a test image of object 2 (hereinafter referred to as the first test image). The first test image includes an image area (hereinafter referred to as the first image area) that presents the second alignment structure 203 corresponding to surface area 21A. Next, processor 15 performs image analysis on the first test image to detect the presentation position of the first image area in the first test image. If the presentation position of the first image area is not located in the center of the first test image, driving assembly 11 fine-tunes the position of object 2 in the first direction D1. After fine-tuning, photosensitive element 13 captures the first test image again for processor 15 to determine whether the presentation position of the first image area is located in the center of the first test image. Conversely, when the presentation position of the first image block is in the center of the first test image, the photosensitive element 13 captures a test image of the surface block 21A of the object 2 under illumination from the light source assembly 12. After capturing the image, the driving element 11 displaces the next surface block 21B of the object 2 in the first direction D1 to the detection position 14, so that the second alignment structure 203 corresponding to the surface block 21B faces the photosensitive element 13. Next, under illumination from the light source assembly 12, the photosensitive element 13 captures another test image of the object 2 (hereinafter referred to as the second test image). This second test image includes an image block (hereinafter referred to as the second image block) that displays the second alignment structure 203 corresponding to the surface block 21B. The processor 15 then performs image analysis on the second test image to detect the presentation position of the second image block in the second test image. If the presentation position of the second image block is not in the center of the second test image, the driving element 11 fine-tunes the position of the object 2 in the first direction D1. After fine-tuning, the photosensitive element 13 captures another second test image for the processor 15 to determine whether the presentation position of the two image blocks is in the center of the second test image. Conversely, if the presentation position of the second image block is in the center of the second test image, the photosensitive element 13 captures a detection image of surface block 21B of the object 2 under illumination from the light source assembly 12. After capturing, the driving assembly 11 displaces the next surface block 21C of the object 2 in the first direction D1 to the detection position 14, so that the second alignment structure 203 corresponding to surface block 21C faces the photosensitive element 13. Next, under illumination from the light source assembly 12, the photosensitive element 13 captures another test image of the object 2 (hereinafter referred to as the third test image). This third test image includes an image block (hereinafter referred to as the third image block) that displays the second alignment structure 203 corresponding to surface block 21C. The processor 15 then performs image analysis on the third test image to detect the presentation position of the third image block in the third test image. When the presentation position of the third image block is not located in the center of the third test image, the driving component 11 will fine-tune the position of the object 2 in the first direction D1 .After fine-tuning, the photosensitive element 13 captures a third test image again for the processor 15 to determine whether the presentation position of the three image blocks is located in the center of the third test image. Conversely, if the presentation position of the third image block is located in the center of the third test image, the photosensitive element 13 captures a detection image of the surface area 21C of the object 2 under the illumination of the light source assembly 12.

[0126] In some embodiments, when the image scanning system needs to capture an image of an object 2 using two different image capture parameters, the image scanning system sequentially executes an image capture process using each image capture parameter. The different image capture parameters may include providing light L1 of different brightnesses to the light source module 12, illuminating the light source module 12 at different light incident angles, or providing light L1 of different spectra to the light source module 12.

[0127] In some embodiments, reference Figure 23 After capturing the detection images of all surface blocks 21A-21C of object 2, processor 15 stitches the detection images corresponding to all surface blocks 21A-21C of object 2 into an object image in the order of capture (step S221) and compares the stitched object image with a preset pattern (step S222). If the object image does not match the preset pattern, processor 15 adjusts the stitching order of the detection images (step S223) and performs the comparison again after the adjustment (step S222). Conversely, if the object image matches the preset pattern, processor 15 obtains the object image of object 2.

[0128] In some embodiments, the image scanning system may further perform an alignment process. After the object 2 is placed on the driving assembly 11, the image scanning system may perform an alignment process to align the object so as to determine the position of the object 2 at which the image is to be captured.

[0129] Reference Figure 24 During the alignment process, the driving assembly 11 continuously rotates the object 2. As the object 2 rotates, the processor 15 detects the first alignment structure 202 of the object 2 via the photosensitive element 13 (step S201) to determine whether the first alignment structure 202 has reached a predetermined position. During the rotation of the object 2, the second alignment structure 203 of the object 2 sequentially faces the photosensitive element 13.

[0130] In some embodiments, the predetermined type may be the relative position of the first alignment structure 202 and / or the brightness relationship of the image blocks of the first alignment structure 202 .

[0131] In one example, as the object 2 rotates, the photosensitive element 13 continuously captures detection images of the object 2, and these detection images include image blocks representing the first alignment structure 202. The processor 15 analyzes each detection image to determine the relative positions of the image blocks representing the first alignment structure 202 within the detection image and / or the brightness relationships among the image blocks representing the first alignment structure 202 within the detection image. For example, if the processor 15 analyzes the detection image and finds that the image blocks representing the first alignment structure 202 are spaced apart and do not overlap, and that the brightness of the image blocks located in the center of the image blocks representing the first alignment structure 202 is brighter than the image blocks located to the sides, the processor 15 determines that the first alignment structure 202 has reached a predetermined configuration. In other words, the predetermined configuration can be determined based on the image characteristics of a specific structure of the object 2.

[0132] When the first alignment structure 202 reaches a predetermined configuration, the processor 15 stops rotating the object (step S202) and proceeds with the object image capture process. Specifically, the processor 15 controls the drive assembly 11 to stop rotating the object 2. Otherwise, the processor 15 continues capturing inspection images and analyzing the imaging position and / or imaging status of the image area of the first alignment structure 202.

[0133] Based on this, the image scanning system can use image analysis to test the appearance and location of specific structures of the object in the image to determine whether the object is aligned, thereby capturing detection images located at the same position on each surface block based on the aligned object.

[0134] In some embodiments, when the image scanning system has an alignment process, after capturing detection images of all surface blocks 21A- 21C of the object 2 , the processor 15 can stitch these captured detection images into an object image of the object 2 in the capture order (step S231 ).

[0135] For example, Figure 21 and Figure 22 Taking the spindle as an example, after the image scanning system completes the image capture process (i.e., repeatedly executing steps S211 to S215), the photosensitive element 13 can capture the detection images MB of all surface blocks 21A to 21C. Here, the processor 15 can stitch the detection images MB of all surface blocks 21A to 21C into the object image IM of the object 2 in the capture order, as shown in FIG. Figure 15 As shown. In this example, the photosensitive element 13 may be a linear photosensitive element. In this case, the detection image MB captured by the photosensitive element 13 can be stitched by the processor 15 without cropping. In some embodiments, the linear photosensitive element may be implemented by a linear image sensor. Specifically, the linear image sensor may have a field of vision (FOV) approaching 0 degrees.

[0136] In another embodiment, the photosensitive element 13 is a two-dimensional photosensitive element. In this case, when the photosensitive element 13 captures the detection image MB of the surface block 21A-21C, the processor 15 captures the middle section MBc of the detection image MB based on the short side of the detection image MB, such as Figure 25 As shown. The processor 15 then stitches the middle regions MBc corresponding to all surface blocks 21A-21C into the object image IM. In some embodiments, the middle region MBc may have a width of, for example, one pixel. In some embodiments, the two-dimensional photosensitive element may be implemented by an area image sensor. The area image sensor may have a field of view of approximately 5 to 30 degrees.

[0137] In some embodiments, the image scanning system may further include a test procedure. In other words, before executing the alignment procedure and the image capture procedure, the image scanning system may first execute the test procedure to confirm that each component (such as the driving component 11, the light source component 12, and the photosensitive element 13) operates normally.

[0138] In the test procedure, refer to Figure 26 , the photosensitive element 13 captures a test image under the illumination of the light source module 12 (step S301). The processor 15 receives the test image captured by the photosensitive element 13 and analyzes the test image (step S302) to determine whether the test image is normal (step S303), and accordingly determines whether to complete the test. If the test image is normal (the judgment result is "yes"), it means that the photosensitive element 13 has captured a normal inspection image in step S301 of the image capture process. At this time, the image scanning system will continue to execute the alignment process (continue to execute step S201) or continue to execute the image capture process (continue to execute step S211).

[0139] If the test image is abnormal (the judgment result is “No”), the image scanning system may execute a calibration procedure (step S305 ).

[0140] In some embodiments, reference Figure 16 and Figure 17 The image scanning system may further include a light source adjustment component 16, and the light source adjustment component 16 is coupled to the light source component 12 and the processor 15. Here, the light source adjustment component 16 can be used to adjust the position of the light source component 12 to change the light incident angle θ.

[0141] In one example, referring to Figure 16 、 Figure 17 and Figure 26, the photosensitive element 13 can capture a surface block currently located at the detection position 14 as a test image (step S301). At this time, the processor 15 analyzes the test image (step S302) to determine whether the average brightness of the test image meets a preset brightness to determine whether the test image is normal (step S303). If the average brightness of the test image does not meet the preset brightness (the judgment result is "no"), it means that the test image is abnormal. For example, when the light incident angle θ of the light source module 12 is not appropriate, the average brightness of the test image will not meet the preset brightness; at this time, the test image will not be able to correctly present the preset surface shape of the object 2 to be inspected.

[0142] During the calibration process, the processor 15 controls the light source adjustment assembly 16 to readjust the position of the light source assembly 12 and reset the light incident angle θ (step S305). After the light source adjustment assembly 16 readjusts the position of the light source assembly 12 (step S305), the light source assembly 12 emits another test light beam having a different light incident angle θ. The processor 15 then controls the photosensitive element 13 to capture an image of a surface area currently located at the detection position 14 based on the new test light beam (step S301), thereby generating another test image. The processor 15 analyzes the new test image (step S302) to determine whether the average brightness of the new test image meets the predetermined brightness (step S303). If the average brightness of the new test image still does not meet the predetermined brightness (the judgment result is "no"), the processor 15 controls the light source adjustment assembly 16 to further readjust the position of the light source assembly 12 and reset the light incident angle θ (step S301) until the average brightness of the test image captured by the photosensitive element 13 meets the predetermined brightness. When the average brightness of the test image meets the preset brightness (the judgment result is “yes”), the image scanning system then executes the subsequent step S201 or S211 to perform the aforementioned alignment process or image capture process.

[0143] In another embodiment, referring to Figure 16 、 Figure 17 and Figure 27, the processor 15 can also determine whether the setting parameters of the photosensitive element 13 are normal based on whether the test image is normal (step S303). If the test image is normal (the judgment result is "yes"), it means that the setting parameters of the photosensitive element 13 are normal, and the image scanning system then executes the subsequent steps S201 or S211 to perform the aforementioned alignment process or image capture process. If the test image is abnormal (the judgment result is "no"), it means that the setting parameters of the photosensitive element 13 are abnormal, and the processor 15 further determines whether the photosensitive element 13 has performed the adjustment operation of its setting parameters (step S304). If the photosensitive element 13 has performed the adjustment operation of its setting parameters (the judgment result is "yes"), the processor 15 generates a warning signal indicating that the photosensitive element 13 is abnormal (step S306). If the photosensitive element 13 has not performed the adjustment operation of its setting parameters (the judgment result is "no"), the image scanning system enters the aforementioned adjustment process (step S305). During the calibration process, the processor 15 causes the photosensitive element 13 to perform a calibration operation based on its set parameters (step S305). After the photosensitive element 13 performs the calibration operation (step S305), the photosensitive element 13 captures another test image (step S301). The processor 15 then determines whether the test image captured after the photosensitive element 13 performs the calibration operation is normal (step S303). If the processor 15 determines that the test image is still abnormal (the judgment result is "no"), the processor 15 then determines in step S304 that the photosensitive element 13 has performed the calibration operation (the judgment result is "yes"), and the processor 15 generates a warning signal indicating that the photosensitive element 13 is abnormal (step S306).

[0144] In some embodiments, the aforementioned setting parameters of the photosensitive element 13 include a sensitivity value, an exposure value, a focal length value, a contrast setting value, or any combination thereof. In some embodiments, the processor 15 can determine whether the average brightness or contrast of the test image meets a preset brightness to determine whether the aforementioned setting parameters are normal. For example, if the average brightness of the test image does not meet the preset brightness, it indicates that any of the aforementioned setting parameters of the photosensitive element 13 is incorrect, causing the average brightness or contrast of the test image to not meet the preset brightness. If the average brightness or contrast of the test image meets the preset brightness, it indicates that all of the aforementioned setting parameters of the photosensitive element 13 are correct.

[0145] In one embodiment, the image scanning system may further include an audio / video display unit. The aforementioned warning signal may include images, sounds, or both, and the audio / video display unit may display the aforementioned warning signal. Furthermore, the image scanning system may also include a network function, and the processor 15 may use the network function to transmit the aforementioned warning signal to the cloud for storage, or transmit the warning signal to other devices via the network function, thereby notifying users of the cloud or other devices of the abnormality of the photosensitive element 13 and enabling them to perform debugging operations on the photosensitive element 13.

[0146] In one embodiment, during the calibration process (step S305), the photosensitive element 13 automatically adjusts its setting parameters according to a parameter setting file. The parameter setting file stores the setting parameters of the photosensitive element 13. In some embodiments, the examiner updates the parameter setting file through the user interface of the image scanning system, causing the photosensitive element 13 to automatically adjust its setting parameters according to the updated parameter setting file during the calibration process to correct any incorrect setting parameters.

[0147] In the aforementioned embodiment, when the photosensitive element 13 captures an image (ie, a test image or a detection image), the light source assembly 12 emits a light beam L1 toward the detection position 14 , and the light beam L1 illuminates the surface area currently located at the detection position 14 in an oblique or sideways direction.

[0148] Reference Figure 18 and Figure 19 The incident direction of light L1 forms an angle (hereinafter referred to as light incident angle θ) with the forward normal 14A of the surface area at the detection location 14. That is, at the light incident end, the angle between the optical axis of light L1 and the forward normal 14A is the light incident angle θ. In some embodiments, the light incident angle θ is greater than 0 degrees and less than or equal to 90 degrees. That is, relative to the forward normal 14A, the detection light L1 illuminates the detection location 14 at a light incident angle θ greater than 0 degrees and less than or equal to 90 degrees, so that the surface area currently at the detection location 14 is illuminated by the detection light L1 from the side or oblique direction.

[0149] In some embodiments, as Figure 18 and Figure 19 As shown, the photosensitive axis 13A of the photosensitive element 13 is parallel to the positive normal 14A; or Figure 20 As shown, the photosensitive axis 13A of the photosensitive element 13 is located between the forward normal line 14A and the extension direction D1. That is, there is an angle (hereinafter referred to as the light reflection angle α) between the photosensitive axis 13A and the forward normal line 14A. The photosensitive element 13 receives diffuse light generated by the surface blocks 21A-21C. The photosensitive element 13 captures detection images of each surface block 21A-21C sequentially located at the detection position 14 based on the diffuse light (step S130 or step S214).

[0150] In some embodiments, based on a light incident angle θ greater than 0 degrees and less than or equal to 90 degrees, that is, based on sideways or obliquely incident light L1, if surface 21 of object 2 includes groove-like or hole-like surface structures, light L1 will not reach the bottom of the surface structures. The surface structures will appear as shadows in the inspection image of surface areas 21A-21C, thus forming an inspection image with a clear contrast between surface 21 and surface defects. In this way, the image scanning system or the inspector can determine whether surface 21 of object 2 has defects by detecting the presence of shadows in the image.

[0151] In some embodiments, according to different light incident angles θ, surface structures with different depths appear with different brightness in the detection image. Figure 19 As shown, when the light incident angle θ is equal to 90 degrees, the incident direction of the light L1 is perpendicular to the depth direction of the surface defect, that is, the optical axis of the light L1 overlaps with the tangent line of the surface at the center of the detection position; at this time, regardless of the depth of the surface structure, the surface structure on the surface 21 does not generate reflected light and diffuse light due to the concave part not being illuminated by the light L1. The deeper or shallower surface structures appear as shadows in the detection image, that is, the detection image has poor contrast, or is close to no contrast. Figure 18 As shown, when the light incident angle θ is less than 90 degrees, the incident direction of the detection light L1 is not perpendicular to the depth direction of the surface structure; at this time, the light L1 irradiates a partial area of the surface structure under the surface 21, and the partial area of the surface structure is irradiated by the light L1 to generate reflected light and diffused light. Therefore, the photosensitive element 13 receives the reflected light and diffused light from the partial area of the surface structure, and the surface structure presents an image with a brighter boundary (such as a convex boundary of a defect) or a darker boundary (such as a concave boundary of a defect) in the detection image, that is, the detection image has a better contrast.

[0152] Moreover, in the case of the same light incident angle θ being less than 90 degrees, the photosensitive element 13 receives more reflected light and diffused light from the shallower surface structure than from the deeper surface structure. Therefore, compared with the surface structure with a larger depth-to-width ratio, the shallower surface structure appears as a brighter image in the detection image. Furthermore, in the case of the light incident angle θ being less than 90 degrees, the smaller the light incident angle θ, the more reflected light and diffused light are generated in the surface structure area, the surface structure appears as a brighter image in the detection image, and the brightness of the shallower surface structure in the detection image is also greater than the brightness of the deeper surface structure in the detection image. For example, compared with the detection image corresponding to the light incident angle θ of 60 degrees, the surface structure appears higher in the detection image corresponding to the light incident angle θ of 30 degrees; and, in the detection image corresponding to the light incident angle θ of 30 degrees, the shallower surface structure appears higher in the detection image than the deeper surface structure.

[0153] Therefore, there is a negative correlation between the magnitude of the light incident angle θ and the brightness of the surface structures in the inspection image. When the light incident angle θ is smaller, shallower surface structures appear brighter in the inspection image. This means that the image scanning system or inspector has a harder time identifying shallower surface structures when the light incident angle θ is smaller. In other words, the image scanning system or inspector can more easily identify deeper surface structures based on darker images. Conversely, when the light incident angle θ is larger, both shallower and deeper surface structures appear darker in the inspection image. This means that the image scanning system or inspector can identify all surface structures when the light incident angle θ is larger.

[0154] Therefore, the image scanning system or the inspector can set the corresponding light incident angle θ based on the predetermined hole depth of the predetermined surface structure to be inspected according to the aforementioned negative correlation. For example, if a deeper predetermined surface defect is to be inspected and a shallower predetermined surface structure is to be inspected, the light source adjustment component 16 can adjust the position of the light source component 12 based on the light incident angle calculated from the aforementioned negative correlation to set a smaller light incident angle θ. The light source adjustment component 16 can also drive the light source component 12 to output the inspection light beam L1, so that the shallower predetermined surface defect appears brighter in the inspection image and the deeper predetermined surface structure appears darker in the image. If both shallower and deeper predetermined surface defects are to be inspected, the light source adjustment component 16 can adjust the position of the light source component 12 based on the light incident angle calculated from the aforementioned negative correlation to set a larger light incident angle θ (e.g., 90 degrees). The light source adjustment component 16 can also drive the light source component 12 to output the inspection light beam L1, so that both the shallower and deeper predetermined surface structures appear as shadows in the image.

[0155] For example, assuming that object 2 is a spindle used in a car seat belt assembly, the aforementioned surface structure may be sand holes or air holes caused by dust or air during the manufacturing process of object 2, or dents or scratches. The sand holes or air holes are deeper than the dents or scratches. If it is desired to detect whether object 2 has sand holes or air holes but not dents or scratches, the light source adjustment component 16 can adjust the position of the light source component 12 based on the light incident angle calculated from the aforementioned negative correlation and set a smaller light incident angle θ. This causes the sand holes or air holes to appear lower in the inspection image, while the dents or scratches to appear higher in the inspection image. This allows the image scanning system or the inspector to quickly identify whether object 2 has sand holes or air holes. If the object 2 is to be inspected for bumps, scratches, sand holes, and air holes, the light source adjustment assembly 16 can adjust the position of the light source assembly 12 according to the light incident angle calculated by the aforementioned negative correlation relationship to set a larger light incident angle θ so that bumps, scratches, sand holes, and air holes appear as shadows in the inspection image.

[0156] In some embodiments, the light incident angle θ may be greater than or equal to a critical angle and less than or equal to 90 degrees to obtain the best target feature capture effect at the wavelength to be detected. Here, the critical angle may be related to the surface morphology to be detected. In one embodiment, the light incident angle θ is related to a predetermined depth ratio of a predetermined surface defect to be detected. Figure 29 For example, in the case where the predetermined surface defect includes a predetermined hole depth d and a predetermined hole radius r, the predetermined hole radius r is the distance between any side surface within the predetermined surface defect and the positive normal 14A. The ratio (r / d) between the predetermined hole radius r and the predetermined hole depth d is the aforementioned depth ratio (r / d), and the critical angle is the arctangent (r / d). Therefore, in step S03, the light source adjustment component 16 can adjust the position of the light source component 12 according to the depth ratio (r / d) of the predetermined surface defect to be detected and set the critical angle of the light incident angle θ to the arctangent (r / d). The light incident angle θ must meet the conditions of being equal to or greater than the arctangent (r / d) and less than or equal to 90 degrees. After adjusting the position of the light source component 12, the light source adjustment component 16 drives the light source component 12 to output the detection light beam L1. In some embodiments, the predetermined hole radius r can be pre-set based on the size of the surface structure of the object 2 to be detected.

[0157] In one embodiment, the processor 15 can calculate the light incident angle θ based on the aforementioned negative correlation and arctangent (d / r). The processor 15 then drives the light source adjustment component 16 to adjust the position of the light source component 12 based on the calculated light incident angle θ.

[0158] In some embodiments, the wavelength of the light L1 provided by the light source assembly 12 may be between 300nm and 3000nm. For example, the wavelength of the light L1 may be in the range of 300nm-600nm, 600nm-900nm, 900nm-1200nm, 1200nm-1500nm, 1500-1800nm, or 1800nm-2100nm. In one example, the light L1 provided by the light source assembly 12 may be visible light. Here, visible light can image surface defects on the surface 21 with a μm scale in the inspection image. In some embodiments, the wavelength of the light L1 may be in the range of 380nm to 780nm, which may be determined according to the material properties of the inspection object and the surface spectral reflectivity requirements. In some embodiments, the light L1 may be any one of visible light sources, such as white light, violet light, blue light, green light, yellow light, orange light, and red light. In one embodiment, the wavelength of white light may be between 380nm and 780nm, the wavelength of violet light may be between 380nm and 450nm, the wavelength of blue light may be between 450nm and 495nm, the wavelength of green light may be between 495nm and 570nm, the wavelength of yellow light may be between 570nm and 590nm, the wavelength of orange light may be between 590nm and 620nm, and the wavelength of red light may be between 620nm and 780nm.

[0159] In some embodiments, the light L1 provided by the light source assembly 12 may be far-infrared light (e.g., with a wavelength in the range of 800nm-3000nm). This allows the detection light to capture sub-micron (e.g., 300nm) surface features on the object 2 in the detection image. In one example, when the light source assembly 12 provides oblique illumination of an object 2 with surface attachments, the far-infrared light can penetrate the attachments and reach the surface of the object 2, allowing the photosensitive element 13 to capture an image of the surface of the object 2 beneath the attachments. In other words, the far-infrared light can penetrate the attachments on the surface of the object 2, allowing the photosensitive element 13 to capture an image of the surface 21 of the object 2. In some embodiments, the wavelength of the far-infrared light is greater than 2μm. In some embodiments, the wavelength of the far-infrared light is greater than the thickness of the attachments. In other words, the wavelength of the far-infrared light can be selected based on the thickness of the attachments to be penetrated. In some embodiments, the wavelength of the far-infrared light can also be selected based on the surface morphology of the object to be detected, enabling image filtering of micron (μm) structures. For example, if the sample surface has elongated micro-scratches or sand holes measuring 1 to 3 μm, but these do not affect product quality, quality control personnel are concerned about structural defects larger than 10 μm. The wavelength of the selected far-infrared light L1 can be selected to be intermediate (e.g., 4 μm) to achieve optimal image microstructure filtering and low-noise image quality without affecting the detection of larger defects. Preferably, the wavelength of the far-infrared light is greater than 3.5 μm. In some embodiments, object 2 is preferably made of metal. In some embodiments, the attachments may be oil stains, dirt, paint, etc.

[0160] In one embodiment, the processor 15 can drive the light source adjustment component 16 to adjust the intensity of the far-infrared light L1 emitted by the light source component 12 to reduce glare, thereby improving the quality of the detection image captured by the photosensitive element 13, thereby obtaining a low-disturbance, penetrating image. For example, the light source adjustment component 16 can reduce the light intensity so that the photosensitive element 13 obtains a detection image with less glare.

[0161] In another embodiment, surface defects of varying depths appear differently in the inspection image based on varying light incident angles θ, and the intensity of the glare generated by the far-infrared light L1 will also vary accordingly. In other words, the processor 15 can drive the light source adjustment component 16 to adjust the light incident angle θ of the far-infrared light L1 emitted by the light source component 12 to effectively reduce glare, thereby improving the quality of the inspection image captured by the photosensitive element 13 and obtaining a low-disturbance, penetrating image.

[0162] In another embodiment, the light source adjustment component 16 can determine the polarization direction of the far-infrared light L1 emitted by the light source component 12, that is, control the light source component 12 to output polarized detection far-infrared light L1, so as to effectively reduce glare, thereby improving the quality of the detection image captured by the photosensitive element 13, and obtaining a low-disturbance penetrating image.

[0163] In some embodiments, the light source adjustment component 16 can be driven by a motor to adjust the incident angle θ of the light source component 12. The drive motor can be a stepper motor.

[0164] In some embodiments, the aforementioned light source adjustment component 16 may include a driving circuit to adjust the light intensity of the light L1 of the light source component 12 by changing the voltage provided to the light source component 12 .

[0165] In some embodiments, reference Figure 30 The image scanning system may also include a polarizer 17. Polarizer 17 is located on the photosensitive axis 13A of the photosensitive element 13 and between the photosensitive element 13 and the detection position 14. The photosensitive element 13 captures an image of the surface of the object 2 through the polarizer 17. Polarizer 17 provides polarization filtering to effectively prevent saturation glare caused by strong infrared light on the photosensitive element 13, thereby improving the quality of the detection image captured by the photosensitive element 13 and obtaining a low-disturbance, penetrating image.

[0166] In some embodiments, the positions of the light source assembly 12 and the photosensitive element 13 can be designed so that the light incident angle θ is not equal to the light reflection angle α, thereby reducing glare and improving the quality of the detection image captured by the photosensitive element 13 to obtain a low-disturbance penetrating image.

[0167] In some embodiments, the light source adjustment assembly 16 can sequentially adjust the position of the light source assembly 12 so that the photosensitive element 13 captures detection images MB of the object 2 at different light incident angles θ. Consequently, the image scanning system can obtain multiple detection images MB for each surface area of the object 2 at different light incident angles θ. In other words, the photosensitive element 13 captures multiple images of the same surface area based on light L1 at different light incident angles θ, thereby obtaining multiple detection images MB of the same surface area.

[0168] In some embodiments, reference Figure 31 and Figure 32, the image scanning system may also include a spectroscopic component 18. The spectroscopic component 18 is located between the photosensitive element 13 and the detection position 14, or it can be said that the spectroscopic component 18 is located between the photosensitive element 13 and the object 2. The spectroscopic component 18 has a plurality of filter areas F1 corresponding to a plurality of spectra respectively. At this time, the light source component 12 provides a multi-spectral light to illuminate the detection position 14. Here, the multi-spectral light has sub-lights of a plurality of spectra. Therefore, by switching the filter areas F1 of the spectroscopic component 18 (that is, these filter areas F1 are respectively shifted to the photosensitive axis 13A of the photosensitive element 13), the photosensitive element 13 captures the detection image MB of the surface block (one of 21A to 21C) located at the detection position 14 through each filter area F1, so as to obtain a plurality of detection images MB of different spectra. That is, when multi-spectral light is irradiated from light source assembly 12 onto object 2 at detection position 14, the surface of object 2 diffuses the multi-spectral light. The diffused light is filtered by any filter region F1 of spectrometer assembly 18 into sub-light beams having a spectrum corresponding to filter region F1, which then enter the sensing area of photosensitive element 13. At this point, the sub-light beams reaching photosensitive element 130 only have a single spectrum (the middle value of the optical band). When the same filter region F1 is aligned with photosensitive axis 13A of photosensitive element 13, driver assembly 11 shifts one surface segment to detection position 14 at a time. After each shift, photosensitive element 13 captures a detection image MB of the surface segment currently at detection position 14, thereby obtaining detection images MB of all surface segments 21A-21C under the same spectrum. The spectrometer 18 then switches to another filter zone F1 aligned with the photosensitive axis 13A of the photosensitive element 13, sequentially shifting the surface segments and capturing the detection image MB for each of the surface segments. This process continues in this manner, resulting in detection images MB with spectra corresponding to each filter zone F1. In other words, the light source assembly 12 can have a wide range of wavelengths. By placing a spectrometer 18 in the light receiving path that allows for a specific wavelength range, the spectrometer 18 can provide the photosensitive element 13 with reflected light of the desired wavelength L1.

[0169] In some embodiments, reference Figure 31 and Figure 32 The image scanning system may further include a displacement component 19. The displacement component 19 is coupled to the spectroscopic component 18. During operation of the image scanning system, the displacement component 19 sequentially moves one of the filter regions F1 of the spectroscopic component 18 to the photosensitive axis 13A of the photosensitive element 13.

[0170] In another embodiment, the light splitting component can be arranged at the light incident end. Figure 33 and Figure 34The image scanning system may also include a spectroscopic component 18'. The spectroscopic component 18' is located between the light source component 12 and the detection position 14, or in other words, between the light source component 12 and the object 2. The spectroscopic component 18' has a plurality of filter areas F1 corresponding to a plurality of spectra. At this time, the light source component 12 provides a multi-spectral light to illuminate the detection position 14 through the spectroscopic component 18'. Here, the multi-spectral light has sub-lights of a plurality of spectra. Therefore, by switching the filter areas F1 of the spectroscopic component 18' (i.e., these filter areas F1 are respectively shifted to the optical axis of the light source component 12), the multi-spectral light output by the light source component 12 is filtered into a sub-light of a single spectrum through the filter areas F1 of the spectroscopic component 18', and then illuminates the object 2 at the detection position 14. At this time, the photosensitive element 13 can capture the detection image MB of the specific spectrum of the surface block (one of 21A to 21C) located at the detection position 14. When the same filter region F1 is aligned with the optical axis of the light source assembly 12, the driving assembly 11 shifts one surface segment at a time to the detection position 14. After each shift, the photosensitive element 13 captures a detection image MB of the surface segment currently located at the detection position 14, thereby obtaining detection images MB for all surface segments 21A-21C under the same spectrum. The spectroscopic assembly 18' then switches to another filter region F1 aligned with the optical axis of the light source assembly 12, and again sequentially shifts the surface segments and captures detection images MB for the surface segments. This process continues, resulting in detection images MB having spectra corresponding to each filter region F1. In other words, the light source assembly 12 can have a wider range of wavelengths. By providing a spectroscopic assembly 18 in the incident light path that allows for a specific wavelength range, light L1 of a predetermined wavelength is provided to illuminate the detection position 14.

[0171] In some embodiments, reference Figure 33 and Figure 34 The image scanning system may further include a displacement component 19'. The displacement component 19' is coupled to the spectroscopic component 18'. During operation of the image scanning system, the displacement component 19' sequentially moves one of the filter regions F1 of the spectroscopic component 18' to the optical axis of the light source component 12.

[0172] In some embodiments, the wavelength band of the multi-spectral light provided by the light source assembly 12 may be between 300 nm and 2100 nm, and the wavelength bands that the multiple filter regions F1 of the spectrometer assembly 18 (18') allow to pass may be any non-overlapping segments between 300 nm and 2100 nm. Here, the wavelength bands that the multiple filter regions F1 of the spectrometer assembly 18 (18') allow to pass may be continuous or discontinuous. For example, when the wavelength band of the multi-spectral light may be between 300 nm and 2100 nm, the wavelength bands that the multiple filter regions F1 of the spectrometer assembly 18 (18') allow to pass may be 300 nm-600 nm, 600 nm-900 nm, 900 nm-1200 nm, 1200 nm-1500 nm, 1500 nm-1800 nm, and 1800 nm-2100 nm. In another example, when the wavelength of the multi-spectral light is between 380 nm and 750 nm, the wavelengths allowed to pass through the plurality of filter regions F1 of the spectrometer 18 (18') may be 380 nm to 450 nm, 495 nm to 570 nm, and 620 nm to 750 nm, respectively. In some embodiments, each of the aforementioned spectra may be represented by a wavelength band of monochromatic light or an intermediate value thereof.

[0173] In some embodiments, the light splitting component 18 ( 18 ′) may be a spectroscope. In some embodiments, the displacement component 19 ( 19 ′) may be implemented by a drive motor, which may be a stepper motor.

[0174] In some embodiments, reference Figure 35 The image scanning system can utilize multiple light-emitting elements 121-123 with different spectra to provide light L1 with multiple spectra. Each light-emitting element 121-123 with a different spectrum is activated sequentially, allowing the photosensitive element 13 to obtain detection images with multiple different spectra. In other words, the light source module 12 includes multiple light-emitting elements 121-123, and these light-emitting elements 121-123 correspond to multiple non-overlapping light bands. In some embodiments, these light bands can be continuous or discontinuous.

[0175] For example, the light source assembly 12 includes a red LED, a blue LED, and a green LED. When the red LED emits light, the photosensitive element 13 can obtain a detection image MB of the red light spectrum. When the blue LED emits light, the photosensitive element 13 can obtain a detection image MB of the blue light spectrum, such as Figure 36 When the green LED emits light, the photosensitive element 13 can obtain the detection image MB of the green light spectrum, as shown in FIG. Figure 37As shown. It can be found that the details presented in the detection image MB under different wavelengths of light are different. For example, the grooves presented in the detection image MB under the blue spectrum are more obvious, while the bumps presented in the detection image MB under the green spectrum are more obvious.

[0176] Based on this, the image scanning system can obtain multiple detection images MB of different spectra for each surface block of the same object 2. In other words, the photosensitive element 13 captures multiple images of the same surface block based on light L1 of different wavelength bands to obtain multiple detection images MB of the same surface block of different spectra.

[0177] In some embodiments, as Figure 28 、 Figure 31 and Figure 33 As shown, the light source assembly 12 may include a light emitting element.

[0178] In some other embodiments, Figure 18 、 Figure 19 and Figure 30 As shown, the light source assembly 12 may include two light-emitting elements 121 and 122, and the two light-emitting elements 121 and 122 are symmetrically arranged on opposite sides of the object 2 relative to the forward normal 14A. The two light-emitting elements 121 and 122 respectively illuminate the detection position 14. The surface 21 is illuminated by the symmetrical detection light L1, generating symmetrical diffuse light. The photosensitive element 13 sequentially captures the detection images of each surface block 21A-21C located at the detection position 14 based on the symmetrical diffuse light, thereby improving the imaging quality of the detection image. In some embodiments, the light-emitting elements 121 and 122 can be implemented by one or more light-emitting diodes (LEDs); in some embodiments, each light-emitting element 121 and 122 can be implemented by a laser light source.

[0179] In one embodiment, the image scanning system may have a single set of light source components 12, such as Figure 14 shown.

[0180] In another embodiment, referring to Figures 38 to 40, the image scanning system may have multiple sets of light source components 12a, 12b, 12c, and 12d. These light source components 12a, 12b, 12c, and 12d are respectively located at different positions of the detection position 14, that is, at different positions of the supporting element 111 that supports the object 2. In this way, the image scanning system is able to obtain an object image with optimal surface feature spatial information. For example, the light source component 12a can be set at the front side of the detection position 14 (or the supporting element 111), the light source component 12b can be set at the back side of the detection position 14 (or the supporting element 111), the light source component 12c can be set at the left side of the detection position 14 (or the supporting element 111), and the light source component 12d can be set at the right side of the detection position 14 (or the supporting element 111).

[0181] Here, under the illumination of each light source component (any of 12a, 12b, 12c, and 12d), the image scanning system performs an image capture process to obtain a detection image MB of all surface blocks 21A to 21C of the object 2 when illuminated from a specific direction. For example, the image scanning system first emits light L1 from light source component 12a. When light source component 12a emits light L1, the photosensitive element 13 captures the detection image MB of all surface blocks 21A to 21C of the object 2. Then, the image scanning system switches to emitting light L1 from light source component 12b. When light source component 12b emits light L1, the photosensitive element 13 also captures the detection image MB of all surface blocks 21A to 21C of the object 2. Next, the image scanning system switches to emitting light L1 from light source component 12c. When light source assembly 12c emits light L1, photosensitive element 13 also captures detection images MB of all surface areas 21A-21C of object 2. The image scanning system then switches to light source assembly 12d emitting light L1. When light source assembly 12d emits light L1, photosensitive element 13 also captures detection images MB of all surface areas 21A-21C of object 2.

[0182] Based on this, the image scanning system can obtain multiple detection images MB of each surface block of the same object 2 under the illumination of different light source modules 12a~12d. In other words, the photosensitive element 13 will perform multiple image captures for the same surface block based on the light L1 provided by different light source modules 12a~12d to obtain multiple detection images MB of the same surface block at different illumination directions. In some embodiments, the image scanning system can integrate the object images IM of the object 2 under different illumination directions to improve the imaging distinction of various surface morphologies of the object, thereby improving the recognition of the surface morphology of the object and obtaining the best resolution of the surface morphology of the object. For example, the image scanning system can integrate the object images IM of the object 2 under different illumination directions to form an imaging effect with obvious differences between the attachments and the surface morphology of the surface 21 of the object 2, so as to facilitate the recognition of the surface morphology of the object.

[0183] For example, refer to Figure 41 The image scanning system may include four light source modules 12a, 12b, 12c, and 12d, which are respectively arranged on the upper side, lower side, left side, and right side of the detection position 14. Assume that the surface 21 of the object 2 has a surface morphology of attachments and slots Sb of the pattern Sa. In the image capture process, the photosensitive element 13 may capture an image M01 (e.g., Figure 42 As shown), an image M02 (as shown) of the surface 21 of the object 2 is captured based on the light L1 provided by the light source module 12b. Figure 43 As shown), an image M03 (as shown) of the surface 21 of the object 2 is captured based on the light L1 provided by the light source module 12a. Figure 44 ), and an image M04 (as shown) of the surface 21 of the object 2 captured based on the light L1 provided by the light source module 12d. Figure 45 See Figures 42 to 45 In the images M01-M04, the image of the pattern Sa will not produce shadows due to different lighting directions, while the image of the slot Sb will produce corresponding shadows due to different lighting directions. The processor 15 will superimpose the images M01-M04 of the object 2 into a superimposed object image (i.e., the initial image IMc), as shown in FIG. Figure 46 When the initial image IMc is fed into the artificial neural network system 30 or any of the sub-neural network systems 33, the artificial neural network system 30 or the sub-neural network system 33 can determine whether the surface of the object 2 has attachments and / or surface morphology based on the shadow expression (e.g., presence or absence, presentation position, etc.) in the initial image IMc. Figure 46 Taking the initial image IMc as an example, the sub-neural network system 33 can determine whether the surface 21 of the object 2 has attachments and slots after executing the prediction model based on the initial image IMc.

[0184] In some embodiments, the optical axes (e.g., light L1) of any two adjacent light source modules 12a-12d have the same predetermined angle. For example, in a top-view image scanning system, the light source modules 12a-12d are arranged around the center of the detection position 14 at fixed angle intervals.

[0185] In some embodiments, the light source modules 12 a - 12 d provide light L1 toward the detection position 14 at the same light incident angle θ.

[0186] In some embodiments, the photosensitive element 13 captures a plurality of detection images MB of the same spectrum for the same surface area based on the light L1 provided by different light source modules 12 a - 12 d .

[0187] In some embodiments, the photosensitive element 13 can also capture multiple detection images MB of different spectra for the same surface area based on the light L1 provided by different light source modules 12a-12d. For example, assume that the image scanning system has four light source modules 12a-12d, respectively disposed above, below, to the left, and to the right of the detection position 14. The photosensitive element 13 captures a detection image MB of a first spectrum for the surface area 12A based on the light L1 provided by the light source module 12a, a detection image MB of a second spectrum for the surface area 12A based on the light L1 provided by the light source module 12b, a detection image MB of a third spectrum for the surface area 12A based on the light L1 provided by the light source module 12c, and a detection image MB of a fourth spectrum for the surface area 12A based on the light L1 provided by the light source module 12d. The first spectrum to the fourth spectrum each belong to different optical wavelength bands.

[0188] In some embodiments, the photosensitive element 13 captures multiple detection images MB of different spectra for the same surface area based on the light L1 provided by each of the different light source modules 12a-12d. For example, taking light source module 12a and surface area 21A as an example, under the illumination of light source module 12a, the photosensitive element 13 can capture detection images MB of different spectra for the same surface area 21A via the light splitting element 18 (18').

[0189] In one embodiment, if Figure 16 、 Figure 36 and Figure 37As shown, the object 2 has a cylindrical shape, such as a spindle. That is, the body 201 of the object 2 is cylindrical. Here, the surface 21 of the object 2 can be the side surface of the body 201 of the object 2, that is, the surface 21 is a cylindrical surface, and the surface 21 has a radian of 2π. Here, the aforementioned first direction D1 can be a clockwise direction or a counterclockwise direction with the long axis of the body of the object 2 as the rotation axis. In some embodiments, one end of the object 2 is a narrower structure than the other end. In one example, referring to Figure 17 、 Figure 24 and Figure 26 The supporting element 111 may be two rollers separated by a predetermined distance, and the drive motor 112 is coupled to the rotating shafts of the two rollers. Here, the predetermined distance is less than the diameter of the object 2 (the minimum diameter of the body). Therefore, the object 2 is movably disposed between the two rollers. Furthermore, when the drive motor 112 rotates the two rollers, the surface friction between the object 2 and the two rollers drives the two rollers, thereby rotating along the first direction D1 of the surface 21, so that a surface area is aligned with the detection position 14. In another example, the supporting element 111 may be a rotating shaft, and the drive motor 112 is coupled to one end of the rotating shaft. In this case, the other end of the rotating shaft has an insert (such as a socket). In this case, the object 2 may be removably inserted into the insert. Furthermore, when the drive motor 112 rotates the rotating shaft, the object 2 is driven by the rotating shaft to rotate along the first direction D1 of the surface 21, so that a surface area is aligned with the detection position 14. In some embodiments, taking the example of surface 21 being divided into nine surface sections 21A-21C, the driving motor 112 drives the supporting element 111 to rotate 40 degrees each time, thereby rotating the object 2 40 degrees along the first direction D1 of the surface 21. In some embodiments, the angle of rotation of the driving motor 112 in step S13 (to fine-tune the position of the object 2) is smaller than the angle of rotation of the driving motor 112 in step S15 (to shift the next surface section to the detection position 14).

[0190] In one embodiment, if Figures 47 to 50As shown, object 2 is plate-shaped. That is, the main body 201 of object 2 has a flat surface. Surface 21 of object 2 (i.e., the plane of main body 201) may be a non-curved surface with a curvature equal to or approaching zero. Here, the aforementioned first direction D1 may be the direction along which any side length (e.g., the long side) of surface 21 of object 2 extends. In one example, support element 111 may be a planar carrier plate, and drive motor 112 may be coupled to a side of the planar carrier plate. In this case, during the image capture process, object 2 may be removably mounted on the planar carrier plate. Drive motor 112 drives the planar carrier plate to move along first direction D1 of surface 21, thereby displacing object 2 so that a surface area is aligned with detection position 14. Drive motor 112 displaces the planar carrier plate a predetermined distance each time, and by repeatedly driving the planar carrier plate to displace, each surface area 21A-21C is sequentially displaced to detection position 14. Here, the predetermined distance is substantially equal to the width of each surface block 21A- 21C along the first direction D1 .

[0191] In some embodiments, the drive motor 112 may be a stepper motor.

[0192] In one embodiment, referring to Figure 39 and Figure 47 The image scanning system may be provided with a single photosensitive element 13, and the photosensitive element 13 captures images of the plurality of surface blocks 21A-21C to obtain a plurality of detection images corresponding to the surface blocks 21A-21C respectively.

[0193] In one example, it is assumed that the object 2 is cylindrical and the image scanning system is provided with a single photosensitive element 13. The photosensitive element 13 can capture images of multiple surface blocks 21A-21C of the main body (i.e., the middle section) of the object 2 to obtain multiple detection images MB corresponding to the surface blocks 21A-21C respectively. The processor 15 then stitches the detection images MB of the surface blocks 21A-21C into the object image IM, as shown in FIG. Figure 15 shown.

[0194] In another embodiment, referring to Figure 16 and Figure 38 The image scanning system may include a plurality of photosensitive elements 13 facing the detection position 14 and arranged along the long axis of the object 2. The photosensitive elements 13 respectively capture detection images of different sections of the surface of the object 2 located at the detection position 14.

[0195] In one example, it is assumed that the object 2 is cylindrical and the image scanning system is provided with a plurality of photosensitive elements 131 to 133, such as Figure 16These photosensitive elements 131-133 respectively capture detection images MB1-MB3 of the surface of the object 2 at different sections of the detection position 14, and then the processor 15 stitches all the detection images MB1-MB3 into the object image IM, as shown. Figure 51 For example, assuming that the number of the photosensitive elements 131-133 is three, and the processor 15 stitches the object image IM of the object 2 according to the detection images MB1-MB3 captured by the three photosensitive elements 131-133, as shown in FIG. Figure 51 The object image IM includes a sub-object image 22 ( ) formed by stitching together the detection images MB1 of all the surface blocks 21A-21C captured by the first photosensitive element 131 of the three photosensitive elements 13. Figure 51 The sub-object image 23 ( Figure 51 The sub-object image 24 ( Figure 51 (the lower part of the object image IM in the).

[0196] In some embodiments, although the aforementioned image scanning system is described by taking the capture of detection images of all surface blocks of the object 2 as an example, the present invention is not limited thereto. The image scanning system can also be used to directly capture detection images of the entire surface of the tiny object 2 (i.e., the surface of the object 2 facing the sensing element 13, and the area of this surface is equal to or smaller than the viewing angle of the sensing element 13), or by setting it to capture detection images of only any one area or any multiple areas among all surface blocks of the object 2.

[0197] In some embodiments, the processor 15 may not execute the stitching process, but directly use the detection image MB of any surface block 21A- 21C of the object 2 captured by the photosensitive element 13 as the object image IM.

[0198] In some embodiments, the processor 15 can automatically determine whether the surface 21 of the object 2 contains surface defects, whether the surface 21 has different textures, and whether the surface 21 has attachments such as paint or oil stains based on the obtained object image IM. That is, the processor 15 can automatically determine the different surface types of the object 2 based on the object image.

[0199] In some embodiments, the processor 15 may include the aforementioned artificial neural network system 30 to automatically classify the surface type of the object image IM, thereby automatically determining the surface type of the surface 21 of the object 2. In one example, before the artificial neural network system 30 is established, the object image IM generated by the processor 15 may be trained (deep learning) by multiple sub-neural network systems 33 of the artificial neural network system 30 to establish respective prediction models for identifying the surface type of the object. In another example, before the artificial neural network system 30 is established, the object image IM generated by the processor 15 may be fed into multiple trained sub-neural network systems 33 to obtain respective determination defect rates. Furthermore, the processor 15 may further connect these sub-neural network systems 33 in series based on their respective determination defect rates to form the artificial neural network system 30. In yet another example, after the artificial neural network system 30 is established, the object image IM generated by the processor 15 may be predicted and classified by the artificial neural network system 30, so that the classification prediction of the object image IM is performed sequentially through the prediction models of each sub-neural network system 33.

[0200] In some embodiments, the object image IM generated by the processor 15 can be fed into another processor having the aforementioned artificial neural network system 30, so that the artificial neural network system 30 can automatically classify the surface type based on the obtained object image IM, thereby automatically determining the surface type of the surface 21 of the object 2.

[0201] In one embodiment, the image scanning system and the artificial neural network system 30 can be implemented on the same host. For example, the image scanning system and the artificial neural network system 30 are implemented on the same host, and the host has a processor 15 to control the operation of the image scanning system and execute the artificial neural network system 30. In another exemplary embodiment, the image scanning system and the artificial neural network system 30 are implemented on the same host, and the host has a processor 15 to control the operation of the image scanning system and another processor to execute the artificial neural network system 30. In another embodiment, the image scanning system and the artificial neural network system 30 can also be implemented on different hosts. In other words, the image scanning system and the artificial neural network system 30 are implemented on two different hosts. These two hosts can be connected in a wired or wireless communication manner to transmit information such as object images IM.

[0202] In some embodiments, the creation and application of the artificial neural network system 30 can be implemented on different processors (or hosts). In other words, after one processor connects multiple sub-neural network systems 33 in series to form the artificial neural network system 30, the formed artificial neural network system 30 is then loaded onto another processor for execution.

[0203] In some embodiments, the aforementioned sub-neural network system 33 can also operate independently. In one embodiment, the processor 15 can have any sub-neural network system 33 to automatically classify the surface morphology based on the obtained object image IM. In the learning phase, the sub-neural network system 33 performs deep learning on the obtained object image IM to establish a prediction model of the object surface morphology. In the prediction phase, the object image IM generated by the processor 15 can be continuously predicted and classified by the sub-neural network system 33 using the prediction model to identify the surface morphology of the object. In some embodiments, the object image IM generated by the processor 15 can be fed into another processor having the aforementioned sub-neural network system 33 so that the sub-neural network system 33 can automatically classify the surface morphology based on the obtained object image IM.

[0204] In one embodiment, the image scanning system and the sub-neural network system 33 can be implemented on the same host. For example, the image scanning system and the sub-neural network system 33 are implemented on the same host, and the host has a processor 15 to control the operation of the image scanning system and execute the sub-neural network system 33. In another exemplary embodiment, the image scanning system and the sub-neural network system 33 are implemented on the same host, and the host has a processor 15 to control the operation of the image scanning system and another processor to execute the sub-neural network system 33. In another embodiment, the image scanning system and the sub-neural network system 33 can also be implemented on different hosts. In other words, the image scanning system and the sub-neural network system 33 are implemented on two different hosts. These two hosts can be connected in a wired or wireless communication manner to transmit information such as object images IM.

[0205] In some embodiments, when a sub-neural network system 33 operates independently, the creation (i.e., the learning phase) and application (i.e., the prediction phase) of the sub-neural network system 33 can be implemented on different processors (or hosts). In other words, one processor has an untrained neural network system 33 and trains the neural network system 33 (performing deep learning) using multiple object images IM to establish its prediction model. The trained sub-neural network system 33 is then loaded onto another processor to perform classification prediction.

[0206] For example, in one example, when object 2 is an unqualified object, its surface will have one or more surface patterns that the artificial neural network system has learned and attempted to capture, enabling at least one sub-neural network system 33 to select it. Conversely, when object 2 is a qualified object, its surface will not have any surface patterns that have been recorded to trigger the selection action of any sub-neural network system 33. During the learning phase, the object image IM received by the sub-neural network system 33 will include a portion of the object image IM labeled with one or more surface patterns, while another portion will include a portion of the object image image IM without any surface pattern. Furthermore, the output of the sub-neural network system 33 will be pre-defined into multiple surface pattern categories based on these surface patterns. In another example, when object 2 is an unqualified object, the surface of object 2 may have one or more first-type surface morphologies that the artificial neural network has learned and attempted to capture. Conversely, when object 2 is a qualified object, the surface of object 2 may have one or more second-type surface morphologies that the artificial neural network has learned and attempted to capture. The second-type surface morphology may be, for example, a standard surface morphology. During the learning phase, the object image IM received by the sub-neural network system 33 may include a portion labeled with one or more first-type surface morphologies and a portion labeled with one or more second-type surface morphologies. Furthermore, the output of the sub-neural network system 33 may be pre-defined into multiple surface morphology categories based on these surface morphologies.

[0207] Reference Figure 52 When the surface of the object 2 has at least one surface morphology, the corresponding image position of the object image IM of the object will also present the partial images P01 - P09 of the surface morphology.

[0208] In some embodiments, during the learning phase, the object image IM received by the sub-neural network system 33 is of known surface type (i.e., the target surface type is already marked thereon), and the surface type category output by the sub-neural network system 33 is also set. In other words, each object image IM used for deep learning is already marked with the existing object type. In some embodiments, the category label of the object type can be presented as a label pattern on the object image IM (e.g., Figure 52 as shown), and / or record the object information in the image information of the object image IM.

[0209] In some embodiments, during the learning phase, the sub-neural network system 33 is trained using an object image IM of known surface morphology to generate judgment items for each neuron in the prediction model and / or adjust the weights of the connections of any neurons so that the prediction results of each object image IM (i.e., the output surface defect category) are consistent with its known and labeled learned surface morphology, thereby establishing a prediction model for identifying the surface morphology of the object.

[0210] During the prediction phase, the sub-neural network system 33 uses the established prediction model to perform classification predictions on object images IM of unknown surface types. In some embodiments, the sub-neural network system 33 predicts the percentage of each surface type category that each object image IM is likely to fall into. The sub-neural network system 33 then determines whether the corresponding object 2 meets the criteria based on the percentage of each surface type category in the object image IM and categorizes the object image IM into either a normal or abnormal group based on the qualified or unqualified status of the object image IM.

[0211] In some embodiments, the processor 15 includes one or more sub-neural network systems 33. During the learning phase, the object image input to each sub-neural network system 33 is a known surface morphology. After the object image with the known surface morphology is input, each sub-neural network system 33 performs deep learning based on the known surface morphology and the surface morphology category of the known surface morphology (hereinafter referred to as the preset surface morphology category) to establish a prediction model (i.e., composed of multiple hidden layers connected in sequence, each hidden layer having one or more neurons, and each neuron performing a judgment item). In other words, during the learning phase, the object image with the known surface morphology is used to generate the judgment item of each neuron and / or adjust the weight of the connection between any two neurons so that the prediction result of each object image (i.e., the output preset surface morphology category) conforms to its known surface morphology.

[0212] For example, assuming the aforementioned surface morphologies can be sand holes, air holes, bumps, or scratches, image blocks representing different surface morphologies can include image blocks with sand holes of varying depths, image blocks with bumps or scratches but no sand holes, image blocks with varying surface roughness, image blocks with no surface defects, image blocks with different contrast ratios resulting from illuminating surface blocks 21A-21C with detection light L1 of varying wavelengths, or image blocks with attachments of varying colors. During the learning phase, the sub-neural network system 33 performs deep learning based on the aforementioned object images of varying surface morphologies to establish a predictive model for identifying various surface morphologies. Furthermore, the sub-neural network system 33 can categorize the object images of varying surface morphologies and pre-generate different preset surface morphology categories. Therefore, during the prediction phase, after the object image IM is input, the artificial neural network system 30 (or each sub-neural network system 33) executes its prediction model based on the input object image to identify the object image that represents the surface morphology of the object 2. The prediction model classifies the object image according to a plurality of predetermined surface morphology categories.

[0213] For example, taking a sub-neural network system 33 as an example, the sub-neural network system 33 executes the aforementioned prediction model based on the object image fed in. The sub-neural network system 33 can use the object image IM of the object 2 to identify that the surface block 21A of the first object 2 includes sand holes and bumps, the surface block 21B of the second object 2 does not have surface defects, the surface block 21C of the third object 2 includes sand holes and paint, and the surface roughness of the surface block 21A is greater than the surface roughness of the surface block 21C; then, the surface morphology category includes sand holes or air bubbles. For example, taking the six categories of surface defects, namely, holes, scratches or bumps, high roughness, low roughness, with attachments, and without surface defects, the sub-neural network system 33 may classify the object image IM of the first object 2 into the preset categories of sand holes or air holes and scratches or bumps, classify the object image IM of the second object 2 into the preset category of without surface defects, classify the object image IM of the first object 3 into the preset category of sand holes or air holes and the preset category of with attachments, and classify the detected image of the object image IM of the third object 2 into the preset category of uneven roughness. In another exemplary embodiment, after the classification, the sub-neural network system 33 may further output the object images IM of the first object 2 and the third object 2 as an abnormal group and output the object image IM of the second object 2 as a normal group based on the classification results.

[0214] In some embodiments, the artificial neural network system 30 or any of the sub-neural network systems 33 according to the present invention can be implemented as a computer program product, such that when a computer (i.e., a processor) loads and executes the program, the method for screening object surface morphology based on an artificial neural network according to any of the aforementioned embodiments can be performed. In some embodiments, the computer program product can be a non-transitory computer-readable recording medium, and the program is stored in the non-transitory computer-readable recording medium for loading by a computer (i.e., its processor). In some embodiments, the program itself can be a computer program product and can be transmitted to a computer via wired or wireless means.

Claims

1. A method for screening surface morphology of objects based on an artificial neural network, suitable for screening multiple objects, characterized in that: include: Performing deep learning on multiple untrained sub-neural network systems under different training conditions to establish multiple trained sub-neural network systems that are not connected to each other; Using the plurality of trained and unconnected sub-neural network systems to perform surface morphology recognition on a plurality of object images of the same batch to perform classification prediction of the plurality of object images and obtain a respective determined defect rate for each of the trained and unconnected sub-neural network systems based on the classification, wherein the plurality of object images correspond to the surface morphology of a portion of the plurality of objects; and The multiple sub-neural network systems that have been trained and are not connected to each other are connected in series into an artificial neural network system based on the judgment defect rate of each sub-neural network system to screen the remaining objects among the multiple objects, wherein the multiple sub-neural network systems are connected in series into an artificial neural network system based on the judgment defect rates of the prediction models of the multiple sub-neural network systems from high to low.

2. The method for screening object surface morphology based on artificial neural network according to claim 1, wherein: Also includes: Convert each object image into a matrix; One of the plurality of sub-neural network systems performs the surface shape recognition using the matrix.

3. The method for screening object surface morphology based on artificial neural network according to claim 1, wherein: Also includes: naturalizing the plurality of object images; and Converting the normalized multiple object images into a matrix; One of the plurality of sub-neural network systems performs the surface shape recognition using each of the matrices.

4. The method for screening object surface morphology based on artificial neural network according to claim 1, wherein: Also includes: superimposing the plurality of object images corresponding to the same object into an initial image; One of the plurality of sub-neural network systems performs the surface shape recognition using each of the initial images.

5. The method for screening object surface morphology based on artificial neural network according to claim 1, wherein: Also includes: superimposing the plurality of object images corresponding to the same object into an initial image; and Convert each of the initial images into a matrix; One of the plurality of sub-neural network systems performs the surface shape recognition using each of the matrices.

6. The method for screening object surface morphology based on artificial neural network according to claim 1, wherein: Each sub-neural network system is implemented using a convolutional neural network (CNN) algorithm.

7. The method for screening object surface morphology based on artificial neural network according to claim 1, wherein: Each object image is formed by stitching together multiple detection images.

8. The method for screening object surface morphology based on artificial neural network according to claim 1, wherein: The multiple sub-neural network systems have different numbers of neural network layers.

9. The method for screening object surface morphology based on artificial neural network according to claim 1, wherein: The plurality of sub-neural network systems have different neuron configurations.

10. The method for screening object surface morphology based on artificial neural network according to claim 1, wherein: Also includes: A plurality of object images corresponding to the remaining objects are fed into the artificial neural network system to perform the surface shape recognition.

11. The method for screening object surface morphology based on artificial neural network according to claim 1, wherein: Also includes: The remaining objects are preferentially screened based on the ones having higher defect rates determined by the sub-neural network systems.

12. The method for screening object surface morphology based on artificial neural network according to claim 1, wherein: The plurality of object images correspond to surface morphologies of the plurality of objects with known defect rates, and one of the determined defect rates of each of the sub-neural network systems is higher than the known defect rate.

Images