Two-dimensional material visual sensor capable of realizing infrared and visible dual-band fusion calculation and image recognition system

Abstract

The invention discloses a two-dimensional material visual sensor capable of realizing infrared and visible dual-band fusion calculation and an image recognition system. The two-dimensional material visual sensor sequentially comprises a top electrode, a first material layer, a second material layer, a third material layer and a bottom electrode from top to bottom, wherein the first material layer, the second material layer, the third material layer and the bottom electrode show a light guide effect; the third material layer generates a photovoltaic effect through a Schottky barrier formed by the third material layer and the bottom electrode, an open-circuit voltage generated by the photovoltaic effect generates a driving effect on the light guide effect of the first material layer, and finally a light current containing infrared and visible information is output; meanwhile, an external bias voltage can be applied to the first material layer, and the output current can be enhanced or suppressed by controlling the direction of the external bias voltage relative to the open-circuit voltage generated by the photovoltaic effect.

Description

Technical Field

[0001] The present invention belongs to the field of photoelectric detection and visual sensing technology, and specifically relates to a two-dimensional material visual sensor and an image recognition system capable of realizing infrared and visible dual-band fusion calculation. Background Art

[0002] Dual-band and multi-band fusion photoelectric detection technology occupies an important position in visual processing tasks and has a wide range of applications. Among them, the fusion processing of infrared light and visible light is one of the most common branches of multi-band fusion processing. With the development of fields such as autonomous driving and atmospheric environment monitoring in recent years, its importance has been further enhanced.

[0003] Infrared light and visible light occupy different positions in the electromagnetic spectrum and have their own unique characteristics. Infrared light has a longer wavelength, usually between 700 nanometers and 10 millimeters, and has the ability to penetrate smoke, haze and thin clouds, so it can work effectively in extreme weather conditions. In addition, infrared light has advantages in detecting thermal radiation and can stably extract contour information, so it is widely used in night vision monitoring and thermal imaging; visible light has a shorter wavelength, between 400 nanometers and 700 nanometers, which is the spectral range visible to the human eye. Visible light has rich color and texture details, and can provide high-resolution and high-contrast image information in imaging.

[0004] Traditional photoelectric detectors can usually only detect light information in a single band, such as infrared or visible light, which often makes it difficult to provide comprehensive and stable image information when dealing with complex and changing environments. In order to solve this problem, dual-band fusion detection technology of infrared and visible light has emerged. This technology combines the advantages of the two bands to provide more comprehensive and accurate image information, significantly improving the detection and computing processing capabilities of the system.

[0005] In recent years, two-dimensional materials are new materials with great potential after silicon-based materials in the above-mentioned field of photoelectric detection, and have many irreplaceable advantages. Two-dimensional materials refer to materials composed of a single layer or a few layers of atoms. Graphene is the first two-dimensional material discovered and studied. Subsequently, scientists discovered a series of other two-dimensional materials, such as transition metal dichalcogenides (TMDs, such as MoS2, WS2), black phosphorus (BP), hexagonal boron nitride (h-BN), etc. These materials have atomic-level thickness, high controllability, and excellent electrical and optical properties, showing great potential in infrared and visible light detection. Its layered structure produces unique electronic structure and optical properties. The response wavelength and photoelectric conversion efficiency can be controlled by means of control such as gate voltage and bias voltage. The prepared photodetector can achieve wide spectrum, high sensitivity, and can operate in room temperature environment.

[0006] In the prior art, dual-band photoelectric detectors mainly integrate infrared sensors and visible light sensors, transmit front-end data backward to the processing unit, and realize information fusion with the help of back-end processing. This solution usually includes four steps: hardware integration, image detection, data transmission, and algorithm fusion. Hardware integration refers to integrating infrared sensors and visible light sensors in the same device, building an optical system through mechanical structure design, and then performing image acquisition, that is, the two sensors collect image information in the infrared band and visible light band respectively, and finally perform data fusion, use algorithms to process data at the back end, fuse the image information of the two bands, and generate the final image. In the field of computer vision, the advancement of deep learning technology has made dual-band image fusion and processing possible. By training neural networks, the system can automatically extract and fuse the feature information of the dual bands, and improve the accuracy of tasks such as image classification, target detection, and semantic segmentation.

[0007] Currently on the market, FLIR Duo Pro R from FLIR Systems and DJI Zenmuse XT2 from DJI both integrate dual-spectrum cameras for visible light and infrared thermal imaging, which are widely used in drone systems. They can capture high-resolution visible light images and infrared thermal images at the same time, and are suitable for surveying and mapping, agricultural monitoring, building inspection, search and rescue, etc. In addition, Sony's IMX500 intelligent vision sensor is the world's first CMOS image sensor with built-in AI processing capabilities, supporting the fusion perception of infrared and visible light.

[0008] Although the above devices have achieved the fusion of visible and infrared dual-band information to a certain extent, they still have the following significant shortcomings:

[0009] (1) Material limitations: In the field of infrared detection, traditional silicon-based materials have their own bandgap limitations (the longest wavelength can reach 1100nm near-infrared), while HgCdTe materials with a wavelength range in the infrared band have high requirements for growth process and require a low-temperature environment during detection. Most existing technologies are based on silicon-based materials and have wavelength and process limitations;

[0010] (2) Pixel alignment problem: In existing technologies, infrared and visible light sensors must achieve precise pixel alignment, otherwise image distortion and information loss will occur. To solve this problem, most products will first use algorithms to perform pixel alignment during back-end processing, which will consume a certain amount of computing power and cause processing delays;

[0011] (3) Resolution and sensitivity differences: There are significant differences in the resolution and sensitivity of infrared sensors and visible light sensors, resulting in inconsistent image quality after fusion, which makes it difficult for the system to provide consistent high-quality images under various environmental conditions;

[0012] (4) Huge consumption of data transmission: Existing technologies often use the method of front-end detection and back-end processing, which requires the transmission of a large amount of data generated by the front-end to the back-end. This data transfer consumes a lot of resources and causes delays.

[0013] (5) Data processing complexity: The fusion of dual-band information at the back end requires efficient algorithms and powerful computing capabilities, such as pixel registration algorithms and fusion algorithms that implement different functions. These algorithms increase the complexity and power consumption of the system. In addition, since a large amount of data processing is required at the back end, the real-time performance and response speed of the system are affected.

[0014] (6) Integration and cost issues: Integrating multiple sensors into a compact system is complex and expensive, and existing technologies find it difficult to strike a balance between cost and performance. Summary of the invention

[0015] Purpose of the invention: In order to solve the problems of material limitation, data handling and back-end processing mentioned in the prior art, the present invention proposes a two-dimensional material visual sensor and image recognition system that can realize infrared and visible dual-band fusion calculation.

[0016] Technical solution: A two-dimensional material visual sensor that can realize infrared and visible dual-band fusion calculations, which includes, from top to bottom: a top electrode, a first material layer exhibiting a photoconductive effect, a second material layer exhibiting a dielectric and optical blocking effect, a third material layer and a bottom electrode; the third material layer generates a photovoltaic effect through a Schottky barrier formed with the bottom electrode, and the open-circuit voltage generated by the photovoltaic effect drives the photoconductive effect of the first material layer, and finally outputs a photocurrent containing infrared and visible information.

[0017] Furthermore, the open circuit voltage generated by the photovoltaic effect produces a driving effect on the photoconductive effect of the first material layer, specifically including:

[0018] Infrared light penetrates the first material layer and reaches the Schottky barrier, and electrons and holes are separated under the electric field of the Schottky barrier, generating an open circuit voltage; the generated open circuit voltage is applied to one end of the first material layer through the common electrode, producing a driving effect on the photoconductive effect of the first material layer.

[0019] Furthermore, the first material layer is one of a MoS2 layer, a graphene layer, a black phosphorus (BP) layer, a WS2 layer, a SnS2 layer, a Bi2Se3 layer and an InSe layer.

[0020] Furthermore, the second material layer is a h-BN layer, S i One of an O2 layer, a HfO2 layer, a ZrO2 layer and an Al2O3 layer.

[0021] Furthermore, the third material layer forms a metal-semiconductor contact with the bottom electrode, and the Schottky barrier of the metal-semiconductor contact generates an open circuit voltage, which can be replaced by: the metal and the semiconductor form other types of Schottky barriers to generate an open circuit voltage, a p-type semiconductor and an n-type semiconductor form a pn junction to generate an open circuit voltage, two semiconductor materials with different band gaps form a heterojunction to generate an open circuit voltage, the same material forms a homojunction through doping or electric field regulation to generate an open circuit voltage, a quantum well and quantum dot structure forms a quantum confinement effect to generate an open circuit voltage, and an organic photovoltaic structure generates an open circuit voltage.

[0022] Furthermore, the third material layer is a PdSe2 layer.

[0023] Furthermore, an external bias voltage is applied to the first material layer to achieve regulation of the output current.

[0024] Furthermore, the step of applying an external bias voltage to the first material layer to adjust the output current specifically includes:

[0025] An external bias is applied to the first material layer, and the output current is enhanced or suppressed by controlling the direction of the external bias relative to the open-circuit voltage generated by the photovoltaic effect.

[0026] The present invention discloses an image recognition system, comprising: a visual sensor and a neural network;

[0027] The visual sensor is used to generate a corresponding photocurrent at each pixel of the image to be identified in the infrared and visible dual bands;

[0028] The neural network is used to use the photocurrent generated at each pixel as an input neuron to obtain an image recognition result.

[0029] Among them, the visual sensor is a two-dimensional material visual sensor disclosed above that can realize infrared and visible dual-band fusion calculation.

[0030] Furthermore, the neural network is a neural network that only includes an input layer and an output layer.

[0031] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0032] (1) The sensor of the present invention couples two different photoelectric response mechanisms through device design. At the detection end, the fusion calculation of infrared and visible band information can be completed by relying on the physical mechanism of the device itself, realizing the process from two light intensity signals (infrared and visible signals) to one current output;

[0033] (2) In the infrared and visible dual-band fusion detection and calculation tasks, the front-end detection and processing scheme has higher energy efficiency and faster response speed than the front-end detection and back-end processing scheme, and is more suitable for functions such as autonomous driving. Existing front-end processing schemes often switch detection of different bands by applying additional voltages, and use cache units to achieve fusion calculations in timing. The sensor of the present invention uses the physical mechanism of the material itself to achieve fusion calculations of visible and infrared bands at the same time and in the same place. Specifically, the sensor of the present invention uses photovoltaic (PV) effect and photoconductive (PC) effect to respond to infrared light and visible light respectively. Through effective driving between materials, the open-circuit voltage generated by the photovoltaic effect of the lower material can drive the photoconductive effect of the upper material, and finally form an output photocurrent containing infrared and visible information, realizing the fusion calculation function of infrared and visible band information.

[0034] (3) The sensor of the present invention can further regulate the enhancement or suppression effect on the output current by applying a bias voltage. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is an optical microscope picture of a two-dimensional material visual sensor disclosed in the present invention that can realize infrared and visible dual-band fusion calculation;

[0036] Figure 2 The material band diagram and photoelectric response mechanism principle diagram of a two-dimensional material visual sensor capable of realizing infrared and visible dual-band fusion calculation disclosed by the present invention, wherein: Figure 2 The left picture shows the photovoltaic effect mechanism of PdSe2 material. Figure 2 The right picture in the figure shows the photoconductive effect mechanism of MoS2 material;

[0037] Figure 3 A longitudinal schematic diagram of a two-dimensional material visual sensor capable of realizing infrared and visible dual-band fusion calculation and its equivalent circuit disclosed in the present invention;

[0038] Figure 4 The photocurrent mapping diagram corresponding to a two-dimensional material visual sensor disclosed in the present invention that can realize infrared and visible dual-band fusion calculation under different visible light intensities, infrared light intensities and bias voltages;

[0039] Figure 5 A flowchart showing a verification example of the present invention;

[0040] Figure 6 The following is a demonstration result of a verification example of the present invention. DETAILED DESCRIPTION

[0041] The technical solution of the present invention is now further described in conjunction with the accompanying drawings and embodiments.

[0042] Embodiment 1:

[0043] In computer vision, there are still challenges such as the complexity of multi-band data fusion algorithms, high computing requirements for real-time processing, and calibration and synchronization of data from different bands. Solving these challenges requires continuous hardware improvements and algorithm optimization to further improve performance and applicability. To this end, this embodiment proposes a two-dimensional material vision sensor that can realize infrared and visible dual-band fusion calculations. The sensor can generate images under an optical microscope, such as Figure 1 As shown by Figure 1 It can be seen that the basic structure of the sensor proposed in this embodiment is a MoS2 / h-BN / PdSe2 three-layer stacked device, which is prepared by transferring palladium diselenide (PdSe2) material, hexagonal boron nitride (h-BN) material and molybdenum disulfide (MoS2) material to a silicon substrate with a bottom electrode using a PVA wet transfer method. The bottom electrode and the top electrode are prepared using electron beam lithography (EBL) and electron beam evaporation (EBE) methods.

[0044] In this embodiment, the main purpose of the molybdenum disulfide (MoS2) layer is to exhibit the photoconductive effect. Other materials that exhibit the photoconductive effect may also replace molybdenum disulfide (MoS2), including but not limited to: graphene, black phosphorus (BP), WS2, SnS2, Bi2Se3 and InSe. It should be noted that the selection of this layer of material needs to match the photovoltaic voltage that can be generated by the Schottky barrier formed by palladium diselenide (PdSe2) and the metal electrode. As long as they can match, the function of the present invention can be achieved. Here, the Schottky barrier formed by palladium diselenide (PdSe2) and the metal electrode is only an example. When palladium diselenide (PdSe2) and the metal electrode are replaced by other materials, they must also match the drivable photovoltaic voltage corresponding to the layer of material.

[0045] In this embodiment, the functions of hexagonal boron nitride (h-BN) include: as a dielectric, isolating molybdenum disulfide (MoS2) and palladium diselenide (PdSe2) to avoid interlayer crosstalk; and as an optical blocking layer, reflecting and scattering a portion of visible light. Therefore, any material that can exhibit dielectric and optical blocking effects can replace hexagonal boron nitride (h-BN), including but not limited to S i O2, HfO2, ZrO2 and Al2O3.

[0046] In this embodiment, palladium diselenide (PdSe2) forms a Schottky barrier with the bottom electrode to generate an open circuit voltage. The open circuit voltage generated here needs to match the material of the first layer (in this embodiment, molybdenum disulfide (MoS2)). This layer of material mainly relies on the heterojunction formed with the metal to generate photovoltaic bias. Therefore, any barrier that can generate photovoltaic bias can be used, such as metals and semiconductors forming other types of Schottky barriers to generate open circuit voltages, p-type semiconductors and n-type semiconductors forming pn junctions to generate open circuit voltages, two semiconductor materials with different band gaps forming heterojunctions to generate open circuit voltages, the same material forming homojunctions through doping or electric field regulation to generate open circuit voltages, quantum wells and quantum dot structures forming quantum confinement effects to generate open circuit voltages, and organic photovoltaic structures generating open circuit voltages. The photovoltaic voltage of this layer is determined by the properties and thickness of the material. The sensor of this embodiment provides a new front-end implementation method for infrared and visible light dual-band fusion calculations. The sensor of this embodiment directly outputs the light intensity signals of infrared and visible light as photocurrent signals at the front end, and can also adjust the output photocurrent by changing the bias voltage to realize dual-band fusion calculation. The realization principle is now described.

[0047] Figure 2 It is shown that in the sensor of this embodiment, the PdSe2 material and the MoS2 material exhibit different photoelectric response mechanisms. Figure 2 The left figure in the figure shows that PdSe2 material mainly relies on the Schottky barrier formed with the bottom electrode to produce photovoltaic effect. Infrared light penetrates the MoS2 material and reaches the Schottky barrier. Electrons and holes are separated under the electric field of the Schottky barrier, generating an open circuit voltage; Figure 2 The right picture in the figure shows that MoS2 material exhibits photoconductive effect.

[0048] Figure 3 The vertical schematic diagram of the sensor of this embodiment and its equivalent circuit are shown. The open circuit voltage generated by the bottom Schottky barrier is applied to one end of the MoS2 material through the common electrode to provide a bias voltage for its photoconductive (PC) effect. This is the embodiment of the invention that introduces an external bias voltage V DS As an additional degree of freedom for the fusion of infrared and visible light, the enhancement and suppression effects are achieved, and this capability can be used for front-end processing of redundant data. In addition, this embodiment uses a common electrode to connect the upper and lower layers, realizing the coupling of the photovoltaic (PV) effect and the photoconductive effect in mechanism, thereby realizing dual-band fusion calculation of infrared and visible light.

[0049] Figure 4 The photocurrent mapping diagram corresponding to the sensor of this embodiment is shown by applying three bias voltages of 20mV, 0mV, and -20mV under different visible light intensities and infrared light intensities.

[0050] The above characteristics enable the sensor of this embodiment to realize many computer vision functions at the front end that can only be completed at the back end.

[0051] This embodiment uses the physical mechanism of the material itself to achieve fusion calculation of visible and infrared bands at the same time and place, that is, it can realize in-situ detection of infrared and visible light, fundamentally solving the problem of alignment of different photoelectric sensors in computer vision. The sensor of this embodiment uses the photovoltaic effect and the photoconductive effect to respond to infrared light and visible light respectively. Through the effective drive between the materials, the open circuit voltage generated by the photovoltaic effect of the lower material can drive the photoconductive effect of the upper material, and finally form an output photocurrent containing infrared and visible information to realize the calculation function. In addition, the sensor of this embodiment can further regulate the enhancement or suppression effect of the output current by applying an external bias voltage.

[0052] Embodiment 2:

[0053] Now, a two-dimensional material visual sensor proposed in Example 1 that can realize infrared and visible dual-band fusion calculation is used for letter recognition. On the one hand, the feasibility of the sensor proposed in Example 1 is verified, and on the other hand, an experimental conclusion is drawn that the sensor proposed in Example 1 can realize front-end fusion calculation. The specific implementation process is as follows:

[0054] Step 1: MoS2 and h-BN flakes were obtained from bulk crystals by mechanical exfoliation method;

[0055] Step 2: The bottom electrode (2 / 25nm Ti / Au) was fabricated on a SiO2 (280nm) / Si substrate by standard electron beam lithography (EBL) and electron beam evaporation (EBE);

[0056] Step 3: Transfer PdSe2, h-BN, and MoS2 layer by layer to the bottom electrode using polyvinyl alcohol (PVA) wet transfer technology.

[0057] Step 4: Deposit the top electrode using electron beam lithography and electron beam evaporation to obtain Figure 1 The MoS2 / h-BN / PdSe2 three-layer stacking structure shown.

[0058] Step 5: Use a common electrode to connect the upper and lower layers to achieve dual-band fusion calculation of infrared and visible light.

[0059] according to Figure 3 The structure shown in the figure is used to observe the current changes of the entire sensor under different infrared and visible light intensities through a self-made automatic control system and corresponding control program, and the photocurrent of the sensor under different light intensities is obtained. ph Used to characterize the photoresponse of the sensor, which is the current I under light conditions lightThe current I under dark conditions dark difference.

[0060] Figure 4 The photocurrent mapping diagram of the sensor is shown under different visible light intensities and infrared light intensities by applying three external bias voltages of 20mV, 0mV, and -20mV. In this embodiment, an external bias voltage V is applied to the non-common end of MoS2. DS , by controlling the external bias voltage V DS The open circuit voltage V generated by the Schottky barrier formed between PdSe2 and the bottom electrode OC direction to achieve enhancement and suppression of the photocurrent signal. Finally, three types of photocurrent data were collected under 36 sets of different light intensities, and all photocurrent data were converted into a two-dimensional current mapping matrix, with infrared light intensity on the x-axis, visible light intensity on the y-axis, and photocurrent represented by grayscale. From the comparison of the three different mapping diagrams, it can be seen that the open circuit voltage V generated by applying an external bias of -20mV and applying an external bias of 20mV and the Schottky barrier formed between PdSe2 and the bottom electrode OC Therefore, compared with the external bias of 0 mV, the photocurrent mapping shows enhancement and inhibition respectively.

[0061] like Figure 5 As shown, in this embodiment, light information from two different bands is input into the sensor, each band contains the letters "NJUPHY", and the pixel size is 16×16. The light intensity of each pixel, including no light and the intensity level of the five corresponding bands, is added with noise according to the Gaussian distribution. According to the photocurrent mapping under different external bias voltages, the sensor generates a corresponding photocurrent signal at each pixel, thereby forming a 16×16 photocurrent matrix. Figure 6 In the figure, for the three letters "NJU", a self-powered mode with an external bias of 0mV is uniformly adopted, resulting in a moderate current intensity in the corresponding photocurrent matrix. For the three letters "PHY", three different external biases of 20mV, 0mV and -20mV were applied respectively. Next, these six letters of 16×16 pixel size are input as 1536 input neurons into a simple neural network with only an input layer and an output layer. Using the BP algorithm, the final output results can be clearly distinguished, and NJU, NJUPHY and PHY can be identified under suppression, no change and enhancement effects respectively. Compared with traditional CMOS image sensors, the sensor of this embodiment realizes in-situ detection and fusion calculation at the detector end, realizing the process from two light intensity signals (infrared and visible signals) to one current output. Through device design, simple processing tasks can be completed at the front end, which is of high value for applications that are sensitive to data and delay.

Claims

1. A two-dimensional material visual sensor capable of realizing infrared and visible dual-band fusion calculation, characterized in that: From top to bottom, it includes: a top electrode, a first material layer exhibiting a photoconductive effect, a second material layer exhibiting a dielectric and optical blocking effect, a third material layer and a bottom electrode; the third material layer generates a photovoltaic effect through a Schottky barrier formed with the bottom electrode, and the open-circuit voltage generated by the photovoltaic effect drives the photoconductive effect of the first material layer, and finally outputs a photocurrent containing infrared and visible information.

2. A two-dimensional material visual sensor capable of realizing infrared and visible dual-band fusion calculation according to claim 1, characterized in that: The open circuit voltage generated by the photovoltaic effect produces a driving effect on the photoconductive effect of the first material layer, specifically including: Infrared light penetrates the first material layer and reaches the Schottky barrier, and electrons and holes are separated under the electric field of the Schottky barrier, generating an open circuit voltage; the generated open circuit voltage is applied to one end of the first material layer through the common electrode, producing a driving effect on the photoconductive effect of the first material layer.

3. The two-dimensional material visual sensor capable of realizing infrared and visible dual-band fusion calculation according to claim 1, characterized in that: The first material layer is one of a MoS2 layer, a graphene layer, a black phosphorus layer, a WS2 layer, a SnS2 layer, a Bi2Se3 layer and an InSe layer.

4. The two-dimensional material visual sensor capable of realizing infrared and visible dual-band fusion calculation according to claim 1, characterized in that: The second material layer is an h-BN layer, S i One of an O2 layer, a HfO2 layer, a ZrO2 layer and an Al2O3 layer.

5. The two-dimensional material visual sensor capable of realizing infrared and visible dual-band fusion calculation according to claim 1, characterized in that: The third material layer forms a metal-semiconductor contact with the bottom electrode. The Schottky barrier of the metal-semiconductor contact generates an open circuit voltage, which can be replaced by: the metal and the semiconductor form other types of Schottky barriers to generate an open circuit voltage, a p-type semiconductor and an n-type semiconductor form a pn junction to generate an open circuit voltage, two semiconductor materials with different band gaps form a heterojunction to generate an open circuit voltage, the same material forms a homojunction through doping or electric field regulation to generate an open circuit voltage, a quantum well and quantum dot structure forms a quantum confinement effect to generate an open circuit voltage, and an organic photovoltaic structure generates an open circuit voltage.

6. The two-dimensional material visual sensor capable of realizing infrared and visible dual-band fusion calculation according to claim 1, characterized in that: The third material layer is a PdSe2 layer.

7. The two-dimensional material visual sensor capable of realizing infrared and visible dual-band fusion calculation according to claim 1, characterized in that: An external bias voltage is applied to the first material layer to achieve regulation of the output current.

8. The two-dimensional material visual sensor capable of realizing infrared and visible dual-band fusion calculation according to claim 7, characterized in that: The step of applying an external bias voltage to the first material layer to achieve regulation of the output current specifically includes: An external bias is applied to the first material layer, and the output current is enhanced or suppressed by controlling the direction of the external bias relative to the open-circuit voltage generated by the photovoltaic effect.

9. An image recognition system, characterized in that: include: Vision sensors and neural networks; The visual sensor is used to generate a corresponding photocurrent at each pixel of the image to be identified in the infrared and visible dual bands; The neural network is used to use the photocurrent generated at each pixel as an input neuron to obtain an image recognition result; Wherein, the visual sensor is a two-dimensional material visual sensor capable of realizing infrared and visible dual-band fusion calculation as described in any one of claims 1 to 8.

10. An image recognition system according to claim 9, characterized in that: The neural network is a neural network that only includes an input layer and an output layer.

Images