Signal processing device, signal processing method, and program

Abstract

This signal processing device (100) comprises: a difference signal generation unit (10) which calculates a difference between an input image (71) and a reconstructed image (72) obtained by reconstructing the input image (71) to generate a base difference image (75); a mask signal generation unit (30) which generates a mask image (74) by considering a difference between a processed input image (73) obtained by subjecting the input image (71) to a predetermined process and the input image (71); and an output signal generation unit (50) which generates an output image (76) by masking the base difference image (75) generated by the difference signal generation unit (10) with the mask image (74) generated by the mask signal generation unit (30).

Description

[Technical Field]

[0001] The present invention relates to a signal processing device, a signal processing method, and a program. [Background technology]

[0002] In recent years, advances in AI (Artificial Intelligence) technology have led to the automation of inspection work by utilizing AI in place of manual visual inspection. By utilizing AI, pass / fail judgments can be made with simple settings, eliminating the need to rely on the judgment of workers and allowing for flexible responses when new defects arise. Furthermore, even in industries where high-mix, low-volume production is common, AI can be applied to multiple products, improving the work of inspection personnel.

[0003] In this way, products that use AI-based defect detection engines to detect defects are being developed, and for example, defect detection using image reconstruction is being implemented. Specifically, an inspection image is input into a reconstructed image generation unit to obtain a reconstructed image, and then the inspection image and the reconstructed image are input into an anomaly score calculation unit to calculate an anomaly score and determine whether there is an abnormality. In this case, performance improvement is required in both the generation of the reconstructed image and the calculation of the anomaly score.

[0004] Here, for example, a method for determining the status of image data is disclosed, in which first output data output by a network based on image data is obtained, and second output data is obtained by an algorithm having an effect different from that of the network based on image data, and status information of the image data is determined based on the first output data and the second output data (see Patent Document 1). [Prior art documents] [Patent documents]

[0005] [Patent Document 1] U.S. Patent No. 10,460,213 Summary of the Invention [Problem to be solved by the invention]

[0006] Conventionally, when the difference between a reconstructed image and an input image on which a specific process has been performed is obtained, erroneous detection may occur, and further performance improvement is required.

[0007] In view of the above problem, the present invention aims to reduce false detections that occur when obtaining the difference between a reconstructed image and an input image in a signal processing device, a signal processing method, and a program. [Means for solving the problem]

[0008] That is, the above-mentioned problems of the present invention are solved by the following configuration. (1) a differential signal generation unit that calculates a difference between an input signal and a reconstructed signal obtained by reconstructing the input signal and generates a differential signal; a mask signal generating unit that generates a mask signal by calculating a difference between a processed input signal obtained by performing a predetermined process on the input signal and the input signal; an output signal generation unit that generates an output signal by masking the differential signal generated by the differential signal generation unit with the mask signal generated by the mask signal generation unit; A signal processing device comprising:

[0009] (2) a signal processing unit that performs predetermined processing on the input signal to generate a processed input signal; The signal processing device according to (1) further comprises:

[0010] (3) The signal processing unit applying a smoothing filter to the input signal; A signal processing device according to (2).

[0011] (4) a score calculation unit that calculates an anomaly score for the output signals of the non-defective and defective products processed by the signal processing unit using a plurality of index parameters; a parameter selection unit that selects the index parameter that has the best classification performance for the non-defective product and the defective product from the index parameters, A signal processing device according to (2) or (3).

[0012] (5) The signal processing unit performing processing on the input signal based on the index parameters selected by the parameter selection unit; A signal processing device according to (4).

[0013] (6) a reconstruction signal generation unit that inputs an input signal to a generation deep learning model and generates the reconstruction signal; The signal processing device according to any one of (1) to (3), further comprising:

[0014] (7) The input signal is The information is composed of at least one of an image, an audio, and an output from a sensor. 7. A signal processing device according to any one of claims 1 to 6.

[0015] (8) The differential signal generation unit calculating a difference between the reconstructed signal and the input signal after correcting the reconstructed signal and the input signal; A signal processing device according to any one of (1) to (7).

[0016] (9) The input signal is an image signal, the difference signal generation unit corrects the reconstructed signal and the input signal, and then calculates a difference between the reconstructed signal and the input signal; (8) A signal processing device according to (8).

[0017] (10) The differential signal generating unit determines the absolute value of the difference between the input signal and the reconstructed signal as the differential. A signal processing device according to any one of (1) to (9).

[0018] (11) The differential signal generating unit sets the square of the difference between the input signal and the reconstructed signal as the differential. A signal processing device according to any one of (1) to (9).

[0019] (12) calculating a difference between an input signal and a reconstructed signal obtained by reconstructing the input signal to generate a difference signal; generating a mask signal by calculating a difference between the input signal and a processed input signal obtained by performing a predetermined process on the input signal; masking the difference signal with the mask signal to generate an output signal; A signal processing method comprising:

[0020] (13) A step of calculating a difference between an input signal and a reconstructed signal obtained by reconstructing the input signal to generate a difference signal; a step of generating a mask signal by calculating a difference between a processed input signal obtained by performing a predetermined process on the input signal and the input signal; masking the difference signal with the mask signal to generate an output signal; A program characterized by causing a computer to execute the above. [Effects of the Invention]

[0021] According to the present invention, in a signal processing device, a signal processing method, and a program, it is possible to reduce false detections that occur when a difference between a reconstructed image and an input image is obtained. [Brief explanation of the drawings]

[0022] [Figure 1] 1 is an explanatory diagram illustrating an example of the main configuration of a signal processing device according to an embodiment of the present invention; [Figure 2] FIG. 2 is a functional block diagram of a signal processing device. [Figure 3] 10 is a flowchart showing the process performed by the signal processing device from calculating the difference between an input signal and a reconstructed signal to generating an output signal. [Figure 4]FIG. 10 is an image diagram illustrating the concept of a differential signal generating unit calculating the difference between an input image and a reconstructed image for an arbitrary member to generate a base differential image. [Figure 5] FIG. 10 is an image diagram illustrating the concept of a mask signal generating unit calculating the difference between a processed input image and an input image for an arbitrary member to generate a mask image. [Figure 6] FIG. 2 is a conceptual diagram of processing by a signal processing device. [Figure 7] FIG. 10 is an image diagram illustrating the concept of an output signal generation unit subtracting a mask image from a base differential image for an arbitrary member to generate an output image. [Figure 8] 10 is an explanatory diagram showing an output image generated by an output signal generating unit and displayed on a display unit. FIG. [Figure 9] 10 is a flowchart showing a process in which a processing parameter selection unit selects an index parameter. [Figure 10A] This is a graph showing the relationship between the anomaly score and the number of good / defective products (part 1). [Figure 10B] This is a graph showing the relationship between the anomaly score and the number of good / defective products (part 2). [Figure 10C] This is a graph showing the relationship between the anomaly score and the number of good / defective products (part 3). [Figure 11] FIG. 10 is a functional block diagram of a signal processing device of a comparative example. [Figure 12] 10 is an example of an output image of a signal processing device of a comparative example. DETAILED DESCRIPTION OF THE INVENTION

[0023] The following describes in detail embodiments of the present invention. Note that the embodiments described below are merely examples for realizing the present invention, and should be appropriately modified or changed depending on the configuration of the device to which the present invention is applied and various conditions, and the present invention is not limited to the following embodiments.

[0024] <Present Embodiment> [Overall configuration of image processing device] 1 is an explanatory diagram illustrating an example of the main configuration of a signal processing device 100 according to this embodiment. The signal processing device 100 according to this embodiment is configured to include a CPU (Central Processing Unit) 110, a storage unit 120, a ROM (Read Only Memory) 130, a RAM (Random Access Memory) 140, an input unit 150, a display unit 160, and a communication unit 170.

[0025] The input unit 150 and the display unit 160 constitute an operation display unit 180. The signal processing device 100 is an information processing device such as a personal computer or a server.

[0026] The CPU 110 is a central processing unit that controls the signal processing device 100, and implements each process (function) shown in Fig. 2 by loading an OS (Operating System) and a control program stored in the storage unit 120 or the ROM 130 into the RAM 140 and executing them. Note that each process implemented by the CPU 110 will be described later with reference to Fig. 2.

[0027] The storage unit 120 is a large-capacity storage device, and is configured, for example, by a hard disk drive, a solid state drive (SSD), a non-volatile memory, etc. The storage unit 120 stores a control program 121.

[0028] The ROM 130 stores the programs executed by the CPU 110, as well as data and parameters used in the execution of these programs.

[0029] The RAM 140 is where the OS and various programs are deployed. Furthermore, the RAM 140 functions as a work area for temporarily storing various programs that are read from the ROM 130 and executable by the CPU 110, as well as input data, output data, parameters, and the like generated by the processing of the CPU 110, during various processes executed and controlled by the CPU 110.

[0030] The input unit 150 is configured to include a keyboard equipped with cursor keys, numeric input keys, various function keys, etc., and a pointing device such as a mouse. The input unit 150 outputs press signals of keys pressed on the keyboard and operation signals of the mouse as input signals to the CPU 110. The CPU 110 executes various processes based on the operation signals from the input unit 150.

[0031] The display unit 160 is configured to include a monitor such as a CRT (Cathode Ray Tube) or LCD (Liquid Crystal Display). The display unit 160 displays various screens in accordance with instructions of a display signal input from the CPU 110. A touch panel display can also be used as the display unit 160 and the input unit 150.

[0032] The communication unit 170 includes a communication interface and communicates with external devices on a network.

[0033] The internal bus 190 connects the components of the signal processing device 100 to one another.

[0034] Next, the function of the CPU 110 of the signal processing device 100 according to this embodiment will be described with reference to FIG.

[0035] Fig. 2 is a functional block diagram showing the functions of the CPU 110 of the signal processing device 100 according to this embodiment. The CPU 110 executes the control program 121 shown in Fig. 1 to embody the differential signal generator 10, the reconstruction signal generator 20, the mask signal generator 30, the signal processor 40, the output signal generator 50, and the processing parameter selector 60 shown in Fig. 2.

[0036] The difference signal generation unit 10 calculates the difference between an input image 71 (input signal) and a reconstructed image 72 (reconstructed signal) obtained by reconstructing the input image 71, thereby generating a base difference image 75 (difference signal). Alternatively, the difference signal generation unit 10 may correct the reconstructed image 72 and the input image 71, and then calculate the difference between the reconstructed image 72 and the input image 71. The correction may be, for example, brightness correction.

[0037] Here, the differential signal generating unit 10 may use the absolute value of the difference between the input image 71 and the reconstructed image 72 as the differential, or may use the square of the difference between the input image 71 and the reconstructed image 72 as the differential.

[0038] The input image 71 is an example of an input signal, and the input signal is composed of at least one of an image, a sound, and an output from a sensor. When the input signal is an image signal, the differential signal generator 10 can correct the reconstructed image 72 and the input image 71, and then calculate the difference between the reconstructed image 72 and the input image 71.

[0039] The reconstruction signal generation unit 20 includes a generative deep learning model equipped with an encoder 21 and a decoder 22. The reconstruction signal generation unit 20 inputs an input image 71 into the generative deep learning model to generate a reconstructed image 72. The encoder 21 is a neural network that converts the input image into a latent signal. The decoder 22 is a neural network that restores and reconstructs the original image from the latent signal. This allows the reconstruction signal generation unit 20 to function as a data reconstruction unit. This generative deep learning model operates as an autoencoder or a variational autoencoder that assumes a probability distribution for the latent variable.

[0040] Here, the reconstructed signal generating unit 20 generates a reconstructed image 72 based on the input image 71 using a generative deep learning model that has been trained using input images of multiple non-defective products related to the same inspection target.

[0041] The signal processing unit 40 generates a processed input image 73 (processed input signal) by performing predetermined processing on the input image 71. The signal processing unit 40 applies an image processing filter such as a Gaussian filter to the input image 71. This allows the CPU 110 of this embodiment to reduce erroneous responses in edge portions and the like using the signal processing unit 40. Note that the image processing filter applied by the signal processing unit 40 is not limited to a Gaussian filter, and a predetermined noise suppression filter or the like may also be applied.

[0042] The mask signal generation unit 30 calculates the difference between the input image 71 and a processed input image 73 (processed input signal) obtained by performing a predetermined process on the input image 71 by the signal processing unit 40, and generates a mask image 74 (mask signal).

[0043] The output signal generation unit 50 subtracts the mask image 74 generated by the mask signal generation unit 30 from the base differential image 75 generated by the differential signal generation unit 10 to generate an output image 76 (output signal).

[0044] Note that the processing of the output signal generation unit 50 is not limited to this, and it is also possible to perform mask processing from the base differential image 75 according to the luminance of the mask image 74. The mask processing adjusts the luminance of the target image according to the luminance of the mask image 74. Here, the lower the luminance of the mask image 74, the closer the luminance of the base differential image 75 is to be masked to white (FFFFFF), and conversely, the higher the luminance of the mask image 74, the more the luminance of the base differential image 75 is left unchanged.

[0045] In this embodiment, by later subtracting the variations due to various processes as a mask image 74, it is possible to prevent an increase in the difference due to the processes applied to the input image 71. Furthermore, since the output signal generation unit 50 performs a final calculation by subtraction, it is possible to suppress responses in areas where variations due to various processes have occurred. Furthermore, when calculating the difference between the unprocessed input image 71 and the reconstructed image 72, no new responses will be generated in areas where no response was detected.

[0046] The processing parameter selection unit 60 includes a score calculation unit 61 and a parameter selection unit 62. The processing parameter selection unit 60 adjusts index parameters of the signal processing unit 40 by using information obtained from the output image 76. This allows the CPU 110 of the signal processing device 100 according to this embodiment to determine the content of processing so that the output signal becomes a signal that is most suitable for the intended function.

[0047] The score calculation unit 61 calculates scores 77 (abnormality scores) for non-defective and defective products using a plurality of index parameters.

[0048] The parameter selection unit 62 selects the index parameter (index parameter) that has the best performance in classifying good products and bad products from among the index parameters. The index parameter 78 selected by the parameter selection unit 62 is input to the signal processing unit 40. This allows the processing parameter selection unit 60 to adjust the index parameter of the signal processing unit 40.

[0049] This allows the signal processing unit 40 to process the input image 71 (input signal) based on the index parameters 78 selected by the parameter selection unit 62.

[0050] [Processing of information processing device] Next, the processing of the CPU 110 of the signal processing device 100 according to this embodiment will be described. 3 is a flowchart showing the process performed by the signal processing device 100 of this embodiment, from calculating the difference between an input signal and a reconstructed signal to generating an output signal. Here, an input image 71 is an example of the input signal, and a reconstructed image 72 is an example of the reconstructed signal.

[0051] The input signal is composed of at least one of an image, a sound, and an output from a sensor. If the input signal is an image signal, the differential signal generator 10 can calculate the difference between the reconstructed image 72 and the input image 71 after correcting the reconstructed image 72 and the input image 71.

[0052] First, as shown in FIG. 3, the CPU 110 of the signal processing device 100 applies a generative deep learning model to an input image 71, which is an input signal, by the reconstruction signal generation unit 20, to generate a reconstruction image 72, which is a reconstruction signal (step S001).

[0053] The CPU 110 causes the difference signal generation unit 10 to calculate the difference between the input image 71 and the reconstructed image 72 to generate a base difference image 75 (difference signal) (step S003). As described above, the difference signal generation unit 10 may correct the reconstructed image 72 and the input image 71, and then calculate the difference between the reconstructed image 72 and the input image 71. The correction may be, for example, brightness correction.

[0054] FIG. 4 is an image diagram showing the concept in which the differential signal generating unit 10 of the CPU 110 of this embodiment calculates the difference in brightness of each pixel between an input image 71 and a reconstructed image 72 for an arbitrary component, and generates a base differential image 75.

[0055] Here, as described above, the differential signal generating unit 10 may use the absolute value of the difference in brightness between each pixel of the input image 71 and the reconstructed image 72 as the difference, or may use the square of the difference in brightness between each pixel of the input image 71 and the reconstructed image 72 as the difference.

[0056] 3, the description will continue. The CPU 110 generates a processed input signal by performing predetermined processing on the input image 71 using the signal processing unit 40 (step S005). Here, the signal processing unit 40 applies, for example, a Gaussian filter (smoothing filter) to the input image 71.

[0057] The CPU 110 calculates the difference between the input image 71 and a processed input image 73 (processed input signal) obtained by performing a predetermined process on the input image 71 using the mask signal generation unit 30, and generates a mask image 74 (mask signal) (step S007).

[0058] FIG. 5 is an image diagram showing the concept of the mask signal generating unit 30 of this embodiment calculating the difference between a processed input image 73 and an input image 71 for an arbitrary component and generating a mask image 74.

[0059] 3, the CPU 110 causes the output signal generation unit 50 to subtract the mask image 74 generated by the mask signal generation unit 30 from the base differential image 75 generated by the differential signal generation unit 10 to generate an output image 76 (output signal) (step S009).

[0060] In addition, the CPU 110 may generate an output image 76 by causing the output signal generation unit 50 to mask the base differential image 75 generated by the differential signal generation unit 10 with the mask image 74 generated by the mask signal generation unit 30.

[0061] 6 is an image diagram showing the concept of processing when the input signal is the input image and the reconstructed signal is the reconstructed image. First, the differential signal generator 10 in FIG. 2 calculates the difference between the input image 71 and the reconstructed image 72 (reconstructed signal) obtained by reconstructing the input image 71 to generate a base differential image 75. The signal processor 40 performs a predetermined process on the input image 71 to generate a processed input image 73.

[0062] The mask signal generation unit 30 calculates the difference between the processed input image 73 and the input image 71 to generate a mask image 74. The output signal generation unit 50 then subtracts the mask image 74 from the base difference image 75, or masks the base difference image 75 with the mask image 74, to generate an output image 76.

[0063] FIG. 7 is an image diagram showing the concept of the output signal generating unit 50 of this embodiment subtracting a mask image 74 from a base differential image 75 for an arbitrary component to generate an output image 76.

[0064] FIG. 8 is an enlarged view of an example of an output image 76 generated by the output signal generating section 50 of this embodiment.

[0065] 8, in the output image 76 generated by the output signal generation unit 50, a defect is detected in an area 82, but the false detection in an edge area 81 is very faint. This allows the signal processing device 100 to reduce false detections that occur when obtaining the difference between the reconstructed image and the input image.

[0066] After outputting the output image 76, the CPU 110 of the signal processing device 100 according to this embodiment ends the processing shown in FIG.

[0067] As described above, the CPU 110 of the signal processing device 100 according to this embodiment executes the control program 121 to embody the difference signal generation unit 10, the mask signal generation unit 30, and the output signal generation unit 50. The difference signal generation unit 10 calculates the difference between the input image 71 and the reconstructed image 72 to generate a base difference image 75. The mask signal generation unit 30 calculates the difference between the processed input image 73 and the input image 71 to generate a mask image 74. The output signal generation unit 50 subtracts the mask image 74 from the base difference image 75 or masks the base difference image 75 with the mask image 74 to generate an output image 76.

[0068] This allows the CPU 110 of the signal processing device 100 according to this embodiment to reduce false detections that occur when the difference between the reconstructed image 72 and the input image 71 is obtained.

[0069] In particular, the signal processing device 100 can reduce false positives that are caused by degradation during reconstruction by using the difference between the input image 71 and a processed input image 73 that has undergone various processes as a mask image 74. By subtracting the mask image 74 from the base difference image 75, the signal processing device 100 prevents new reactions from appearing in areas where no reaction occurred, and can suppress an increase in false positives caused by various processes.

[0070] In this way, the CPU 110 of the signal processing device 100 according to this embodiment can reduce false detections due to deterioration during reconstruction by using the difference between the processed input image 73 processed by the signal processing unit 40 and the input image 71 as the mask image 74. Furthermore, the CPU 110 can reduce false detections due to various processes by subtracting the mask image 74 from the base difference image 75 indicating the difference between the input image 71 and the reconstructed image 72.

[0071] Furthermore, as described above, the CPU 110 may generate an output image 76 by causing the output signal generating unit 50 to mask the base differential image 75 generated by the differential signal generating unit 10 with the mask image 74 generated by the mask signal generating unit 30.

[0072] <Processing parameter selection section> The CPU 110 of the signal processing device 100 according to this embodiment can further embody a processing parameter selection unit 60. This allows the CPU 110 of the signal processing device 100 to adjust the content of the processing of the signal processing unit 40 so as to best achieve the objective. In other words, the CPU 110 of the signal processing device 100 according to this embodiment can determine the content of the processing so that the output signal becomes a signal that is most suitable for the intended function.

[0073] The processing parameter selection unit 60 is configured to include a score calculation unit 61 and a parameter selection unit 62 .

[0074] The score calculation unit 61 calculates scores 77 (anomaly scores) of the output signals of the non-defective and defective products processed using a plurality of index parameters by the signal processing unit 40. The parameter selection unit 62 selects, from the plurality of index parameters, the one (index parameter) that has the best performance for classifying non-defective and defective products.

[0075] FIG. 9 is a flowchart showing the process in which the CPU 110 of this embodiment selects the index parameter 78 using the processing parameter selection unit 60.

[0076] As shown in FIG. 9, the score calculation unit 61 of the processing parameter selection unit 60 calculates scores 77 for non-defective and defective products using a plurality of index parameters based on the output image 76 (step S021).

[0077] The parameter selection unit 62 of the processing parameter selection unit 60 selects, from among the index parameters, the index parameter 78 that has the best performance in classifying non-defective and defective products based on the score 77 (abnormality score) (step S023).

[0078] The horizontal axis of the graphs in Figures 10A, 10B, and 10C is the normalized anomaly score. The vertical axis of the graphs is the number of good and bad products. The graphs show the distribution of the number of good and bad products against the anomaly score value.

[0079] 10A, 10B, and 10C, the dashed lines indicate the distribution of the number of non-defective products, and the solid lines indicate the distribution of the number of defective products. Figures 10A, 10B, and 10C each show the evaluation results of a large number of non-defective products and defective products with different index parameters 78. That is, the parameter selection unit 62 evaluates a large number of non-defective products and defective products using multiple parameters, and selects index parameters that can classify these non-defective products and defective products with the highest accuracy.

[0080] 10A, 10B, and 10C, it can be seen that the classification performance for good and defective products changes depending on the index parameter 78. For example, in the graph of FIG. 10A, good and defective products are mixed in a predetermined area. On the other hand, in the graph of FIG. 10B, the mixed area of ​​good and defective products is wide, and the classification performance is lower than in the case of FIG. 10A. On the other hand, in the graph of FIG. 10C, there is almost no mixed area of ​​good and defective products, and good and defective products can be classified with higher accuracy than in the cases of FIGS. 10A and 10B. In this case, the parameter selection unit 62 selects the index parameter 78 applied to the graph of FIG. 10C.

[0081] In this way, when the purpose is anomaly detection, the parameter selection unit 62 can select index parameters 78 that can accurately classify products into good and bad. Note that the parameter selection unit 62 may select index parameters 78 according to the desired purpose, such as index parameters 78 that reduce the false positive rate or index parameters 78 that increase the correct answer rate.

[0082] As described above, the CPU 110 of the signal processing device 100 according to this embodiment can embody the processing parameter selection unit 60, and can therefore adjust the processing content of the signal processing unit 40 so as to best achieve the objective. In other words, the CPU 110 of the signal processing device 100 according to this embodiment can determine the processing content so that the output signal becomes a signal that is most suitable for the intended function.

[0083] <<Operation of Comparative Example>>

[0084] FIG. 11 is a functional block diagram of a signal processing device of a comparative example.

[0085] 11, the signal processing device executes a control program to embody a differential signal generation unit 10 and a reconstruction signal generation unit 20. The reconstruction signal generation unit 20 inputs an input image 71 to a generative deep learning model to generate a reconstructed image 72. The reconstruction signal generation unit 20 is configured to include a generative deep learning model equipped with an encoder 21 and a decoder 22.

[0086] The differential signal generating unit 10 calculates the difference between an input image 71 (input signal) and a reconstructed image 72 (reconstructed signal) obtained by reconstructing the input image 71, and generates an output image 76a.

[0087] FIG. 12 is an enlarged view of an example of an output image of the signal processing device of the comparative example.

[0088] 12, an abnormal region 83 is detected in the output image 76a of the comparative example. However, in this output image 76a, an erroneous response occurs in a region 84, which is simply an edge portion where the luminance change is large.

[0089] In contrast to this, in the output image 76 of this embodiment shown in FIG. 8, an abnormal region 82 is detected, but there are very few false reactions to edge portions where the luminance changes are simply large.

[0090] As described above, the signal processing device 100 of this embodiment can reduce false detections that occur when the difference between a reconstructed image and an input image is obtained. [Explanation of symbols]

[0091] 10 Differential signal generation section 20 Reconstruction signal generation section 21 Encoder (part of a generative deep learning model) 22 Decoder (part of a generative deep learning model) 30 Mask signal generator 40 Signal Processing Unit 50 Output signal generation unit 60 Processing parameter selection section 61 Score calculation section 62 Parameter selection section 71 input images 72 Reconstructed images 73 Processed input image 74 Mask Images 75 base difference images 76,76a Output image 77 score (abnormal score) 78 Index Parameters 81, 82, 83 area 100 Signal processing device

Claims

1. a differential signal generation unit that calculates a difference between an input signal and a reconstructed signal obtained by reconstructing the input signal, and generates a differential signal; a mask signal generating unit that generates a mask signal by calculating a difference between a processed input signal obtained by performing a predetermined process on the input signal and the input signal; an output signal generation unit that generates an output signal by masking the differential signal generated by the differential signal generation unit with the mask signal generated by the mask signal generation unit; A signal processing device comprising:

2. a signal processing unit that performs predetermined processing on the input signal to generate a processed input signal; The signal processing device according to claim 1 , further comprising:

3. The signal processing unit applying a smoothing filter to the input signal; The signal processing device according to claim 2 .

4. a score calculation unit that calculates an abnormality score for the output signals of a non-defective product and a defective product that have been processed by the signal processing unit using a plurality of index parameters; a parameter selection unit that selects the index parameter that has the best classification performance for the non-defective product and the defective product from the index parameters, 4. The signal processing device according to claim 2 or 3.

5. The signal processing unit performing processing on the input signal based on the index parameters selected by the parameter selection unit; The signal processing device according to claim 4 .

6. a reconstructed signal generation unit that inputs the input signal to a generation deep learning model and generates the reconstructed signal; The signal processing device according to claim 1 , further comprising:

7. The input signal is The information is configured by at least one of an image, an audio, and an output by a sensor.

7. A signal processing device according to claim 1.

8. The differential signal generation unit calculating a difference between the reconstructed signal and the input signal after correcting the reconstructed signal and the input signal; 8. A signal processing device according to any one of claims 1 to 7.

9. the input signal is an image signal, the difference signal generation unit corrects the reconstructed signal and the input signal, and then calculates a difference between the reconstructed signal and the input signal; The signal processing device according to claim 8 .

10. the differential signal generation unit determines the absolute value of the difference between the input signal and the reconstructed signal as the differential signal. The signal processing device according to any one of claims 1 to 9.

11. the differential signal generation unit uses the square of the difference between the input signal and the reconstructed signal as the differential signal; The signal processing device according to any one of claims 1 to 9.

12. calculating a difference between an input signal and a reconstructed signal obtained by reconstructing the input signal to generate a difference signal; generating a mask signal by calculating a difference between the input signal and a processed input signal obtained by performing a predetermined process on the input signal; masking the difference signal with the mask signal to generate an output signal; A signal processing method comprising:

13. a step of calculating a difference between an input signal and a reconstructed signal obtained by reconstructing the input signal to generate a difference signal; a step of generating a mask signal by calculating a difference between a processed input signal obtained by performing a predetermined process on the input signal and the input signal; masking the difference signal with the mask signal to generate an output signal; A program characterized by causing a computer to execute the above.

Images