Artificial vision device and operation method thereof

Abstract

An artificial vision device, according to one embodiment of the present invention, comprises: an image generation device configured to generate an image corresponding to an object; a stimulation control device configured to divide the image into a plurality of regions and generate stimulation current information on the basis of the plurality of regions; and a stimulation generation device comprising a plurality of electrodes and configured to apply a current to the plurality of electrodes on the basis of the stimulation current information. The stimulation control device comprises: an object type classification unit configured to generate object type information by classifying the type of the object of the image; a feature extraction unit configured to generate a feature extraction image by dividing the image into the plurality of regions on the basis of the image and the object type information; and a stimulation pattern generation unit configured to generate the stimulation current information which includes information on the magnitudes of the current corresponding to the respective plurality of regions, wherein the magnitudes of the current corresponding to the respective plurality of regions are different from each other.

Description

Artificial visual device and method of operation thereof

[0001] The present invention relates to an artificial vision device and a method of operating the same.

[0002] There are no treatments for retinal degeneration and age-related macular degeneration, which cause vision loss, and to date, artificial vision devices are known to be the only way to improve vision. Artificial vision devices can enable the user to recognize objects by applying electrical stimulation to the retina through multiple electrodes.

[0003] When the same amount of current is applied to each of the electrodes of an artificial vision device, the electrical stimulation may spread to surrounding nerve cells. Consequently, the accuracy of the electrical stimulation applied to the user's retina is reduced, and visual information is distorted, making it difficult for the user to recognize objects.

[0004] The problem that the present invention aims to solve is to provide an artificial vision device having improved performance and a method of operating the same.

[0005] An artificial visual device according to an embodiment of the present invention comprises an image generating device configured to generate an image corresponding to an object, a stimulation control device configured to divide the image into a plurality of regions and generate stimulation current information based on the plurality of regions, and a stimulation generating device configured to apply current to the plurality of electrodes based on the stimulation current information, wherein the stimulation control device comprises an object type classification unit configured to classify the type of the object in the image and generate object type information, a feature extraction unit configured to divide the image into the plurality of regions based on the image and the object type information and generate a feature extraction image, and a stimulation pattern generating unit configured to generate stimulation current information including information on the magnitude of the current corresponding to each of the plurality of regions, wherein the magnitude of the current corresponding to each of the plurality of regions is different from each other.

[0006] In one embodiment, the plurality of regions includes a first region and a second region, and the importance of the information included in the first region is higher than the importance of the information included in the second region.

[0007] In one embodiment, the importance is higher the more important the information is for classifying the type of the object.

[0008] In one embodiment, the plurality of electrodes includes first electrodes corresponding to the first region and second electrodes corresponding to the second region, and the magnitude of the current applied to the first electrodes is greater than the magnitude of the current applied to the second electrodes.

[0009] In one embodiment, the feature extraction unit is further configured to divide the image into the plurality of regions based on explainable artificial intelligence (XAI).

[0010] In one embodiment, the XAI is implemented based on any one of Grad-CAM (Gradient-weighted Class Activation Mapping), LIME (Local Interpretable Model-agnostic Explanations), and RISE (Randomized Input Sampling for Explanation).

[0011] In one embodiment, the object type classification unit is further configured to classify the type of the object based on a pre-trained AI (Artificial Intelligence) model.

[0012] In one embodiment, the image generating device includes an image sensor or a camera device.

[0013] A method of operation of an artificial vision device according to an embodiment of the present invention comprises the steps of: generating an image corresponding to an object; classifying the type of the object based on the image; dividing the image into a plurality of regions based on the image and the type of the object to generate a feature extraction image; generating stimulation current information including information on the magnitude of the current for each of the plurality of regions based on the feature extraction image; and generating an electrical stimulation by applying a current to a plurality of electrodes based on the stimulation current information, wherein the plurality of regions include a first region and a second region, and the plurality of electrodes include first electrodes corresponding to the first region and second electrodes corresponding to the second region, and the magnitude of the first current applied to the first electrodes is greater than the magnitude of the second current applied to the second electrodes.

[0014] In one embodiment, the importance of the information included in the first region is higher than the importance of the information included in the second region.

[0015] In one embodiment, the importance is higher the more important the information is for classifying the type of the object.

[0016] In one embodiment, the step of generating a feature extraction image by dividing the image into multiple regions is performed based on explainable artificial intelligence (XAI).

[0017] In one embodiment, the XAI is implemented based on any one of Grad-CAM (Gradient-weighted Class Activation Mapping), LIME (Local Interpretable Model-agnostic Explanations), and RISE (Randomized Input Sampling for Explanation).

[0018] In one embodiment, the step of classifying the type of the object is performed based on pre-trained Artificial Intelligence (AI).

[0019] According to the present invention, an artificial vision device can apply a larger current to electrodes that apply electrical stimulation necessary for object recognition compared to other electrodes among a plurality of electrodes. Accordingly, unnecessary neural stimulation applied to the user of the artificial vision device can be minimized. In addition, the object recognition performance of the artificial vision device can be improved. Therefore, according to the present invention, an artificial vision device having improved performance and a method of operating the same can be provided.

[0020] FIG. 1 is a block diagram showing an artificial vision device according to an embodiment of the present invention.

[0021] Figure 2 is a block diagram showing the stimulation control device of Figure 1 in more detail.

[0022] Figure 3 is a diagram illustrating the image and feature extraction image of Figure 2.

[0023] Figure 4 is a block diagram showing the stimulus generating device of Figure 1.

[0024] Figure 5 is a diagram illustrating an example of the operation of the artificial vision device of Figure 1.

[0025] Figure 6 is a graph illustrating the power efficiency of the artificial vision device of Figure 1.

[0026] Figure 7 is a graph illustrating the current applied to the electrodes of the electrode array of Figure 4.

[0027] Figure 8 is a graph illustrating the object recognition accuracy of the artificial vision device of Figure 1.

[0028] FIG. 9 is a flowchart for explaining the operation method of the artificial vision device of FIG. 1.

[0029] FIG. 1 is a drawing showing the best mode for carrying out the present invention.

[0030] In the following, embodiments of the present invention will be described clearly and in detail so that a person skilled in the art can easily practice the present invention.

[0031] Terms such as "block," "unit or part," "logic," etc. used in the detailed description, or functional blocks illustrated in the drawings corresponding to them, may be implemented in the form of software, hardware, or a combination thereof.

[0032] FIG. 1 is a block diagram showing an artificial vision device according to an embodiment of the present invention. Referring to FIG. 1, the artificial vision device (100) may include an image generating device (110), a stimulus control device (120), and a stimulus generating device (130).

[0033] The image generating device (110) can generate an image (IMG) corresponding to an external object. In one embodiment, the image generating device (110) may include an image sensor or a camera device. The image generating device (110) can generate an image (IMG) corresponding to an external object by capturing an external object through the image sensor or camera device.

[0034] The stimulus control device (120) can receive an image (IMG) from the image generation device (110). The stimulus control device (120) can classify the type of object in the image (IMG). Specifically, for example, the stimulus control device (120) can determine whether the object is a human, an animal, or an object. For example, if the object is an animal, the stimulus control device (120) can determine what kind of animal the object is. For example, if the object is an object, the stimulus control device (120) can determine what kind of object the object is. In one embodiment, the stimulus control device (120) can classify the type of object based on a pre-trained AI (Artificial Intelligence) model.

[0035] The stimulus control device (120) can divide an image (IMG) into multiple regions based on the type of classified object. The stimulus control device (120) can divide multiple regions based on the importance of information included in a specific region of the image (IMG). For example, the multiple regions may include an important feature region, an intermediate feature region, and a general feature region. For example, the importance of information included in the important feature region may be higher than the importance of information included in the intermediate feature region, and the importance of information included in the intermediate feature region may be higher than the importance of information included in the general feature region. Meanwhile, the importance may be determined based on how important the information included in the corresponding region is for determining the type of object in the image (IMG). For example, the information included in the first region of the image (IMG) may include key information for determining the type of object in the image (IMG). In this case, the stimulus control device (120) can divide the first region into an important feature region.

[0036] In one embodiment, the stimulus control device (120) can divide an image (IMG) into multiple regions through a model based on XAI (Explainable Artificial Intelligence) technology. XAI technology may refer to a technology that explains the decision process or prediction results of an AI model in a way that humans can understand, in order to help understand the basis related to the results output from the AI ​​model. XAI may include a technology for grasping the thinking of an AI model and a technology for expressing the content of the AI ​​model's thinking in a way that is easy for humans to understand.

[0037] Meanwhile, the stimulation control device (120) can generate stimulation current information (info_si). The stimulation current information (info_si) may include information regarding the magnitude of the current applied to each of the plurality of regions. In one embodiment, the stimulation current information (info_si) may include information that the magnitude of the first current corresponding to the important feature region is greater than the magnitude of the second current corresponding to the intermediate feature region, and the magnitude of the second current corresponding to the intermediate feature region is greater than the magnitude of the third current corresponding to the general feature region.

[0038] The stimulation generating device (130) can apply electrical stimulation to the retina of a user of an artificial vision device (100). The stimulation generating device (130) may include a plurality of electrodes that apply electrical stimulation to the retina of a user. The stimulation generating device (130) can apply electrical stimulation to the retina of a user of an artificial vision device (100) by applying current to the plurality of electrodes based on stimulation current information (info_si). Meanwhile, the plurality of electrodes may correspond to a plurality of regions of an image (IMG). For example, the plurality of electrodes may include first electrodes corresponding to important feature regions, second electrodes corresponding to intermediate feature regions, and third electrodes corresponding to general feature regions. In one embodiment, the stimulation generating device (130) may apply a first current to the first electrodes, apply a second current to the second electrodes, and apply a third current to the third electrodes based on stimulation current information (info_si). At this time, the first current is a higher current than the second current, and the second current may be a higher current than the third current.

[0039] Unlike the above description, for example, the artificial vision device (100) may apply the same current to the electrodes without distinguishing the importance of each region of the image (IMG). In this case, an unnecessarily large current may be applied to the electrodes corresponding to regions of the image (IMG) that contain information that is not important for object recognition. In this case, as the electrical stimulation spreads, the user of the artificial vision device (100) may not be able to accurately recognize the object. Additionally, the power efficiency of the artificial vision device (100) may decrease.

[0040] On the other hand, according to an embodiment of the present invention, the artificial vision device (100) can apply currents of different magnitudes to the electrodes. The artificial vision device (100) can apply a strong current to the electrodes corresponding to the area containing information important for object recognition among the electrodes that apply electrical stimulation to the user, and apply a weak current to the electrodes corresponding to the area containing information that is not important. Accordingly, the diffusion of electrical stimulation is reduced, so that the user of the artificial vision device (100) can accurately recognize the object. In addition, the power efficiency of the artificial vision device (100) can be improved.

[0041] FIG. 2 is a block diagram showing the stimulus control device of FIG. 1 in more detail. Referring to FIG. 2, the stimulus control device (120) may include an object type classification unit (121), a feature extraction unit (122), a stimulus information generation unit (123), and a control unit (124).

[0042] The object type classification unit (121) can receive an image (IMG) from an image generation device (110). The object type classification unit (121) can classify the type of object in the image (IMG). The object type classification unit (121) can determine whether the object in the image (IMG) is a human, an animal, or an object. For example, if the object in the image (IMG) is an animal, the object type classification unit (121) can determine what kind of animal the object is. For example, if the object in the image (IMG) is an object, the object type classification unit (121) can determine what kind of object the object is. In one embodiment, the object type classification unit (121) can classify the type of object based on a pre-trained AI model, using object type information (info_CL) that includes information about the type of object in the image (IMG).

[0043] The feature extraction unit (122) can receive an image (IMG) from an image generation device (110) and can receive object type information (info_CL) from an object type classification unit (121). The feature extraction unit (122) can generate a feature extraction image (IMG_ce) by dividing the image (IMG) into multiple regions based on the object type information (info_CL). The feature extraction unit (122) can divide the image (IMG) into multiple regions based on the importance of the information included in the regions of the image (IMG). The feature extraction unit (122) can generate a feature extraction image (IMG_ce) by dividing the image (IMG) into an important feature region containing information of high importance, an intermediate feature region containing information of medium importance, and a general feature region containing information of general importance.

[0044] In one embodiment, the feature extraction unit (122) can generate a feature extraction image (IMG_ce) through an XAI-based model. In one embodiment, XAI can be implemented based on any one of Grad-CAM (Gradient-weighted Class Activation Mapping), LIME (Local Interpretable Model-agnostic Explanations), and RISE (Randomized Input Sampling for Explanation). However, the present invention is not limited thereto, and XAI can be implemented using various algorithms or methods.

[0045] For example, Grad-CAM can determine the importance of stored information by region of an image by analyzing activation maps and gradient information for specific classes. For example, LIME can determine the importance of stored information by region of an image by generating samples in which specific regions of an input image are modified or removed, and then analyzing the impact of these modifications on decisions made by the model. For example, RISE can determine the importance of stored information by region of an image by generating samples in which various regions of an input image are randomly masked, and then statistically analyzing how important the AI ​​model evaluates a region for each sample.

[0046] The stimulation information generation unit (123) can receive a feature extraction image (IMG_ce). The stimulation information generation unit (123) can generate stimulation current information (info_si) that includes information about the magnitude of the current in each region of the feature extraction image (IMG_ce). The stimulation current information (info_si) may include information that the important feature region of the feature extraction image (IMG_ce) corresponds to a first current, the intermediate feature region corresponds to a second current smaller than the first current, and the general feature region corresponds to a third current smaller than the second current.

[0047] The control unit (124) can control the general operation of the components (121~123) of the stimulation control device (120).

[0048] FIG. 3 is a diagram for explaining the image and feature extraction image of FIG. 2. Referring to FIG. 3, an image (IMG) generated by an image generation device (110) may include a first region (R1), a second region (R2), and a third region (R3). Each of the first region (R1), the second region (R2), and the third region (R3) may include information related to the image (IMG) object.

[0049] For example, the information included in the first region (R1) may contain information important for determining the type of object in the image (IMG). For example, the important information may be information related to the unique characteristics of a specific type of object. That is, the information included in the first region (R1) may have high importance. Meanwhile, the information included in the second region (R2) may have medium importance, and the information included in the third region (R3) may have general importance.

[0050] The feature extraction unit (122) can divide the image (IMG) into multiple regions by determining the importance of each region of the image (IMG) based on the importance of the information contained in the regions (R1~R3). Specifically, the feature extraction unit (122) can generate a feature extraction image (IMG_ce) by dividing the image (IMG) into an important feature region (Ric) containing information with high importance, an intermediate feature region (Rmc) containing information with medium importance, and a general feature region (Rnc) containing information with general importance.

[0051] FIG. 4 is a block diagram showing the stimulation generating device of FIG. 1. Referring to FIG. 1 through FIG. 4, the stimulation generating device (130) may include a current control circuit (131) and an electrode array (132).

[0052] The current control circuit (131) can receive stimulation current information (info_si) from the stimulation information generation unit (123). The stimulation information generation unit (123) can apply stimulation current (I_sti) to the electrodes (ED) of the electrode array (132) based on the stimulation current information (info_si).

[0053] The electrode array (132) may include a plurality of electrodes (ED). Each of the plurality of electrodes (ED) may apply electrical stimulation to the retina of the user of the artificial vision device (100). The plurality of electrodes (ED) may include first electrodes (ED1), second electrodes (ED2), and third electrodes (ED3). The first electrodes (ED1) may generate electrical stimulation corresponding to an important feature region (Ric), the second electrodes (ED2) may generate electrical stimulation corresponding to an intermediate feature region (Rmc), and the third electrodes (ED3) may generate electrical stimulation corresponding to a general feature region (Rnc). FIG. 4 illustrates an electrode array including 36 electrodes (ED), but the present invention is not limited thereto, and the number of electrodes (ED) may be varied.

[0054] In one embodiment, the stimulation current information (info_si) may include information that the important feature region (Ric) corresponds to a first current, the intermediate feature region (Rmc) corresponds to a second current which is lower than the first current, and the general feature region (Rnc) corresponds to a third current which is lower than the second current.

[0055] In this case, the current control circuit (131) can apply a first current as a stimulation current (I_sti) to the first electrodes (ED1), apply a second current as a stimulation current (I_sti) to the second electrodes (ED2), and apply a third current as a stimulation current (I_sti) to the third electrodes (ED3). That is, the current control circuit (131) can apply currents having different magnitudes depending on the importance of each region of the image.

[0056] Accordingly, the power efficiency of the artificial vision device (100) can be increased compared to the case where the same current is applied to all electrodes (ED).

[0057] FIG. 5 is a diagram illustrating an example of the operation of the artificial vision device of FIG. 1. Referring to FIGS. 1 through 4, an image generating device (110) can generate an image (IMG) by capturing an external object. An object type classification unit (121) can determine the type of an object in the image (IMG) based on a pre-trained AI model. For example, the object type classification unit (121) can determine that the object in the image (IMG) is a “deer.”

[0058] The feature extraction unit (122) can generate a feature extraction image (IMG_ce) by dividing the image (IMG) into multiple regions based on the image (IMG) and object type information (info_CL). In the feature extraction image (IMG_ce), the region marked in red may be an important feature region (Ric). In the feature extraction image (IMG_ce), the region marked in green may be an intermediate feature region (Rmc). In the feature extraction image (IMG_ce), the region marked in blue may be a general feature region (Rnc).

[0059] The feature extraction unit (122) can determine an area containing information that has high importance in determining the object of the image (IMG) as a “deer” among the areas of the image (IMG) as an important feature area (Ric). The feature extraction unit (122) can determine an area containing information that has medium importance in determining the object of the image (IMG) as a “deer” among the areas of the image (IMG) as an intermediate feature area (Rmc). An area containing information that has general importance in determining the object of the image (IMG) as a “deer” among the areas of the image (IMG) as a general feature area (Rnc).

[0060] Accordingly, for example, the feature extraction unit (122) can generate a feature extraction image (IMG_ce) by combining an important feature region image (IMG_Ric) containing only information of an important feature region (Ric), an intermediate feature region image (IMG_Rmc) containing only information of an intermediate feature region (Rmc), and an intermediate feature region image (IMG_Rnc) containing only information of a general feature region (Rnc).

[0061] For example, a Key Feature Region (Ric) may contain information related to key features indicating that an object in the image is a “deer.”

[0062] The stimulation information generation unit (123) can generate stimulation current information (info_si) including information that the important feature region (Ric) of the feature extraction image (IMG_ce) corresponds to the first current, the intermediate feature region (Rmc) corresponds to the second current which is lower than the first current, and the general feature region (Rnc) corresponds to the third current which is lower than the second current.

[0063] The current control circuit (131) can apply a first current to the first electrodes (ED1) corresponding to the important feature region (Ric) based on the stimulation current information (info_si), apply a second current to the second electrodes (ED2) corresponding to the intermediate feature region (Rmc), and apply a third current to the third electrodes (ED3) corresponding to the general feature region (Rnc).

[0064] Accordingly, electrical stimulation can be applied to the retina of the user of the artificial vision device (100). Accordingly, the user of the artificial vision device (100) can recognize the final image (IMG_f).

[0065] FIG. 6 is a graph for explaining the power efficiency of the artificial vision device of FIG. 1. In the graph of FIG. 6, the horizontal axis represents the number of electrodes included in the electrode array (132), and the vertical axis represents the power consumption of the artificial vision device (100).

[0066] Meanwhile, in FIG. 6, the first method refers to a method of applying the same current to all electrodes of the electrode array (132). The second method refers to a method of dividing an image into multiple regions according to importance, as described above with reference to FIG. 1 to 5, based on an AI model and XAI implemented based on Grad-CAM, and applying a lower current to regions of lower importance. The third method refers to a method of dividing an image into multiple regions according to importance, as described above with reference to FIG. 1 to 5, based on an AI model and XAI implemented based on LIME, and applying a lower current to regions of lower importance. The fourth method refers to a method of dividing an image into multiple regions according to importance, as described above with reference to FIG. 1 to 5, based on an AI model and XAI implemented based on RISE, and applying a lower current to regions of lower importance.

[0067] Referring to FIG. 6, according to an embodiment of the present invention, when the magnitude of the current applied to the electrodes is varied according to importance (i.e., according to the first to third methods), the power consumption of the artificial vision device (100) can be reduced compared to when the same magnitude of current is applied to all electrodes.

[0068] FIG. 7 is a graph for explaining the current applied to the electrodes of the electrode array of FIG. 4. In the graph of FIG. 7, the horizontal axis represents the number of electrodes included in the electrode array (132), and the vertical axis represents the electrode ratio. Referring to FIG. 1 to FIG. 7, as described above, when the magnitude of the current applied to the electrodes is varied according to importance, a high current may be applied to about 20% of the electrodes included in the electrode array (132), a medium current may be applied to 20% of the electrodes, and a low current may be applied to 60% of the electrodes. Accordingly, the power efficiency of the artificial vision device (100) may be increased.

[0069] FIG. 8 is a graph for explaining the object recognition accuracy of the artificial vision device of FIG. 1. In the graph of FIG. 8, the horizontal axis represents the number of electrodes included in the electrode array (132), and the vertical axis represents the object recognition accuracy of the artificial vision device (100). In FIG. 8, the simple recognition method refers to a method of applying the same amount of current to the electrodes of the electrode array (132) without distinguishing the regions of the image (IMG) according to importance. In FIG. 8, the XAI-based recognition method refers to a method of dividing the image into multiple regions according to importance, as described above with reference to FIG. 1 to FIG. 5, and applying a lower amount of current to regions of lower importance. Referring to FIG. 8, even without applying the same amount of current to all electrodes, the user of the artificial vision device (100) can recognize objects with an accuracy similar to that of the simple recognition method. That is, according to an embodiment of the present invention, the power efficiency of the artificial vision device (100) can be increased without reducing the accuracy of object recognition.

[0070] FIG. 9 is a flowchart for explaining the operation method of the artificial vision device of FIG. 1. Referring to FIG. 9, in step S110, the artificial vision device (100) can generate an image (IMG) corresponding to an external object. For example, the image generating device (110) can generate the image (IMG) through a camera or an image sensor.

[0071] In step S120, the artificial visual device (100) can classify the type of object in the image. For example, the stimulus control device (120) can classify the type of object based on a pre-trained AI model.

[0072] In step S130, the artificial visual device (100) can generate a feature extraction image (IMG_ce) based on the type of the image and the classified object. For example, the stimulus control device (120) can generate the feature extraction image (IMG_ce) by dividing the image into multiple regions based on whether the information contained in the image is important information for determining the type of object (i.e., importance).

[0073] In step S140, the artificial visual device (100) can generate stimulation current information (info_sti) based on the feature extraction image (IMG_ce). For example, the stimulation control device (120) can generate stimulation current information (info_sti) that includes information on the current magnitudes by region of the feature extraction image (IMG_ce).

[0074] In step S150, the artificial vision device (100) can generate electrical stimulation based on stimulation current information (info_sti). For example, the stimulation generating device (130) can generate electrical stimulation by applying a high current to electrodes corresponding to important feature regions (Ric), applying a medium current to electrodes corresponding to intermediate feature regions (Rmc), and applying a low current to electrodes corresponding to general feature regions (Rnc), based on stimulation current information (info_sti). Meanwhile, the user of the artificial vision device (100) can recognize a final image (IMG_f) corresponding to an external object by the electrical stimulation applied to the retina.

[0075] The above description describes specific embodiments for implementing the present invention. The present invention will include not only the embodiments described above, but also embodiments that can be simply modified or easily modified. Furthermore, the present invention will include technologies that can be easily modified and implemented using the embodiments. Accordingly, the scope of the present invention should not be limited to the embodiments described above, but should be defined by the claims set forth below as well as equivalents to the claims of this invention.

[0076] The present invention relates to an artificial vision device and a method of operating the same. More specifically, the present invention is applicable to an artificial vision device and a method of operating the same that provides electrical stimulation to the retina of a user of the artificial vision device through an explainable artificial intelligence-based model.

Claims

1. In an artificial visual device, An image generation device configured to generate an image corresponding to an object; A stimulation control device configured to divide the above image into a plurality of regions and generate stimulation current information based on the plurality of regions; and A stimulation generating device comprising a plurality of electrodes and configured to apply current to the plurality of electrodes based on stimulation current information, The above stimulation control device is, An object type classification unit configured to classify the type of the object of the above image and generate object type information; A feature extraction unit configured to generate a feature extraction image by dividing the image into the plurality of regions based on the image and the object type information; and It includes a stimulation pattern generation unit configured to generate stimulation current information including information on the magnitude of the current corresponding to each of the plurality of regions, and An artificial vision device in which the magnitude of the current corresponding to each of the plurality of regions is different.

2. In Paragraph 1, The above plurality of regions are, Includes a first region and a second region, An artificial vision device in which the importance of the information included in the first area is higher than the importance of the information included in the second area.

3. In Paragraph 2, An artificial vision device in which the importance is higher the more important the information is for classifying the type of the object.

4. In Paragraph 2, The above plurality of electrodes are, It includes first electrodes corresponding to the first region and second electrodes corresponding to the second region, An artificial vision device in which the magnitude of the current applied to the first electrodes is greater than the magnitude of the current applied to the second electrodes.

5. In Paragraph 1, The above feature extraction unit is an artificial vision device further configured to divide the image into the plurality of regions based on explainable artificial intelligence (XAI).

6. In Paragraph 5, The above XAI is an artificial vision device implemented based on any one of Grad-CAM (Gradient-weighted Class Activation Mapping), LIME (Local Interpretable Model-agnostic Explanations), and RISE (Randomized Input Sampling for Explanation).

7. In Paragraph 1, The object type classification unit is an artificial vision device further configured to classify the type of the object based on a pre-trained AI (Artificial Intelligence) model.

8. In Paragraph 1, The above image generating device is an artificial vision device comprising an image sensor or a camera device.

9. In the method of operating an artificial vision device, Step of generating an image corresponding to an object; A step of classifying the type of the object based on the image above; A step of generating a feature extraction image by dividing the image into multiple regions based on the type of the image and the object; A step of generating stimulation current information including information on the magnitude of the current for each of the plurality of regions based on the above feature extraction image; and The method includes the step of generating an electrical stimulus by applying current to a plurality of electrodes based on the above stimulation current information, The above plurality of regions include a first region and a second region, and The plurality of electrodes includes first electrodes corresponding to the first region and second electrodes corresponding to the second region, and A method of operation in which the magnitude of the first current applied to the first electrodes is greater than the magnitude of the second current applied to the second electrodes.

10. In Paragraph 9, A method of operation in which the importance of the information included in the first area is higher than the importance of the information included in the second area.

11. In Paragraph 10, The above importance is a method of operation in which the more important the above information is for classifying the above type of object, the higher the importance.

12. In Paragraph 9, The step of generating a feature extraction image by dividing the above image into multiple regions is a method of operation performed based on explainable artificial intelligence (XAI).

13. In Paragraph 11, The above XAI is an operation method implemented based on any one of Grad-CAM (Gradient-weighted Class Activation Mapping), LIME (Local Interpretable Model-agnostic Explanations), and RISE (Randomized Input Sampling for Explanation).

14. In Paragraph 9, The step of classifying the type of the above object is an operation method performed based on pre-trained AI (Artificial Intelligence).

Images