Sensor, manufacturing method thereof and real-time braille-to-voice recognition system

Abstract

The invention discloses a sensor, a manufacturing method thereof and a real-time braille-to-voice recognition system, and the sensor comprises a double-gradient pore structure conductive film which is formed by the electrostatic spinning of a thermoplastic polyurethane elastomer and the treatment of an ultrasonic multi-walled carbon nanotube and a polydimethylsiloxane solution; the polyimide film is of a double-layer structure, and the conductive film of the double-gradient pore structure is located in the double-layer structure of the polyimide film so as to protect the conductive film of the double-gradient pore structure; and the interdigital electrode is in direct contact with the conductive film with the double-gradient pore structure, and the interdigital electrode is located in the double-layer structure of the polyimide film so as to protect the interdigital electrode. The sensor has the advantages of high sensitivity and large linear range, and can realize high-accuracy real-time Braille-to-voice conversion after being applied to a real-time Braille-to-voice recognition system.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensing and monitoring, specifically to a sensor, its manufacturing method, and a real-time Braille-to-speech recognition system. Background Technology

[0002] With the rapid development of the Internet of Things (IoT) and artificial intelligence (AI) technologies, global information technology is evolving towards the interconnection and intelligence of everything. In fields such as next-generation electronic information, new materials, biomedicine, and health, intelligent flexible sensors, as core components for information acquisition and key interfaces for interaction with the environment, are experiencing rapid development and have become a global research frontier. To adapt to increasingly complex special signals and measurement environments, sensor technology is continuously evolving towards intelligence, integration, and miniaturization.

[0003] Compared to traditional silicon-based electronic devices, flexible electronic devices exhibit significant advantages in withstanding mechanical deformations such as compression, stretching, and folding. When applied to wearable devices, flexible sensors can conformally fit the human skin, possessing not only excellent deformation adaptability but also greatly reducing physical irritation and discomfort caused to the skin during long-term wear, while ensuring the accuracy and reliability of signal acquisition, thus significantly improving the user experience. Integrating flexible sensing technology with machine learning algorithms can further expand its application depth and breadth in fields such as health monitoring and assisted diagnosis, achieving interdisciplinary technological integration and performance enhancement.

[0004] According to the World Health Organization, approximately 2.2 billion people worldwide suffer from visual impairment or blindness. Braille, as an indispensable written communication tool for the visually impaired, plays a crucial role in their learning, reading, and writing. However, learning Braille has a high barrier to entry; mastering fluent reading skills usually requires long-term specialized training. For visually impaired individuals using prosthetics or other special circumstances, this process is even more difficult and inconvenient. Therefore, developing efficient and user-friendly Braille assistive reading technology has become an urgent social need.

[0005] Currently, most existing Braille recognition technologies rely on manual feature point extraction, which is inefficient and inconsistent. Furthermore, existing technologies primarily focus on tactile feedback mechanisms, lacking integration with auditory feedback. Compared to traditional recognition methods based on precise point-by-point pressure, the sliding recognition mode, which simulates natural reading behavior, does not require precise positioning of each Braille dot, resulting in higher recognition efficiency and better aligning with the actual usage habits of visually impaired individuals. If machine learning algorithms can be combined to directly convert the signals collected by sensors in real time into speech output, real-time recognition and voice broadcast of Braille information can be achieved, creating a more natural and fluid human-computer interaction channel.

[0006] In conclusion, developing a high-precision, practical real-time Braille-to-speech recognition system will not only help promote the digital preservation and management of historical Braille documents, but also has significant value in improving the mental health of visually impaired individuals and promoting their educational equity and social integration. This technological breakthrough is expected to provide visually impaired communities with more convenient and efficient ways to access and communicate information, thereby substantially improving their quality of life and social participation. Summary of the Invention

[0007] In order to overcome the shortcomings of the prior art, one of the objectives of this invention is to provide a sensor with the advantages of high sensitivity and large linear range. When applied to a real-time Braille-to-speech recognition system, it can achieve high-accuracy real-time Braille-to-speech conversion.

[0008] The second objective of this invention is to provide a method for manufacturing a sensor, which has the advantages of high sensitivity and a large linear range. When the sensor is applied to a real-time Braille-to-speech recognition system, it can achieve high-accuracy real-time Braille-to-speech conversion.

[0009] The third objective of this invention is to provide a real-time Braille-to-speech recognition system, which has the advantages of high sensitivity and a large linear range, and can achieve high-accuracy real-time Braille-to-speech conversion.

[0010] To achieve one of the objectives of this invention, the following solution is adopted: A sensor, comprising: A conductive film with a dual gradient pore structure is formed by electrospinning thermoplastic polyurethane elastomer and then treating it with ultrasonic multi-walled carbon nanotubes and polydimethylsiloxane solution. A polyimide film, wherein the polyimide film has a double-layer structure, and the dual-gradient pore structure conductive film is located in the double-layer structure of the polyimide film to protect the dual-gradient pore structure conductive film. Interdigitated electrodes are in direct contact with the dual-gradient porous conductive film. The interdigitated electrodes are located in the bilayer structure of the polyimide film to protect them.

[0011] Furthermore, the dual-gradient porous structure conductive film is formed by adjusting the electrospinning process parameters.

[0012] Furthermore, the polyimide film is provided with a rectangular positioning frame for positioning the dual-gradient porous conductive film and the interdigitated electrode.

[0013] Furthermore, the dimensions of both the dual-gradient porous conductive film and the interdigitated electrode are 3mm × 3mm.

[0014] Furthermore, it also includes an encapsulation structure for encapsulating the edges of the polyimide film.

[0015] Furthermore, the encapsulation structure is a high-temperature adhesive tape.

[0016] To achieve the second objective of this invention, the following solution is adopted: A method for manufacturing a sensor as described in one of the objectives of this invention includes the following steps: (1) Preparation of polyimide film: Dissolve 4',4-diaminodiphenyl ether in N-methylpyrrolidone, add pyromellitic dianhydride, spin-coat and bake to obtain polyimide film; (2) Fabrication of gradient porous TPU film: TPU and PVA are respectively made into precursor solutions, and after electrospinning, curing and PVA removal, a gradient porous TPU nanofiber film is obtained. (3) Fabrication of a conductive film with a dual gradient pore structure: The TPU nanofiber membrane is immersed in a MWCNTs suspension after ultrasonic treatment, and after cleaning, it is immersed in a hexane solution containing PDMS. After drying, a conductive film composed of MWCNTs, PDMS and TPU is obtained. (4) Fabrication of flexible interdigitated electrodes: Photoresist is spin-coated onto the polyimide film, and after exposure, development, plasma treatment, magnetron sputtering deposition of gold film, and removal of photoresist, flexible interdigitated electrodes are obtained; (5) The dual-gradient porous structure conductive film and the interdigitated electrode are placed in the polyimide film for encapsulation.

[0017] To achieve the third objective of this invention, the following solution is adopted: A real-time Braille-to-speech recognition system includes: A sensing module, comprising a sensor as described in one of the objectives of the present invention, wherein the sensor module is used to acquire analog electrical signals generated when a finger slides or presses on Braille dots via the sensor array; A signal processing module is used to convert the analog electrical signals acquired by the sensing module into digital signals; A data transmission module is used to wirelessly transmit the digital signal obtained by the signal processing module to a mobile device via Bluetooth Low Energy. A character recognition module, configured in the mobile device, is used to input the digital signal transmitted by the data transmission module into a trained convolutional neural network, recognize and output the corresponding Braille character sequence; A speech synthesis module, configured in the mobile device, is used to input the Braille character sequence output by the character recognition module into a text-speech synthesis neural network to generate and output the corresponding speech signal.

[0018] Furthermore, the convolutional neural network in the character recognition module includes 5 convolutional layers, 4 max-pooling layers, and 2 fully connected layers, with a convolutional kernel size of 4, a stride of 1, and a max-pooling layer size of 2.

[0019] Furthermore, the text-to-speech synthesis neural network in the speech synthesis module is a FastSpeech2 neural network, employing a transformer-based decoder.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The sensor of this invention employs a conductive thin film with a "dual-gradient pore structure" as the sensing unit and optimizes its integration with interdigital electrodes, achieving a two-stage deformation amplification effect in its structure. This innovative design successfully solves the problems of low ductility, large size, and inconvenience for wearable integration of traditional rigid sensors. It enables the sensor to maintain excellent flexibility while achieving a significantly expanded linear detection range (0-200 kPa) and extremely high sensitivity (78.41 kPa⁻¹), thus enabling more accurate and reliable capture of subtle and dynamic pressure signals generated when a finger slides across Braille.

[0021] 2. The sensor fabrication method provided by this invention, from the synthesis of polyimide film, electrospinning of gradient-pore TPU film, preparation of MWCNTs / PDMS composite conductive film, to photolithography of flexible interdigitated electrodes, and finally integration and packaging, has clearly defined process parameters and tight connections between each step. This method ensures the controllable fabrication of the core feature of "dual-gradient pore structure," providing process assurance for the reliability and consistency of sensor performance, and is conducive to the large-scale fabrication and application of this high-performance sensor.

[0022] 3. The real-time Braille-to-speech recognition system of this invention deeply integrates the aforementioned high-performance sensor array, high-precision signal acquisition circuit, low-power wireless transmission module, and advanced machine learning algorithms (CNN and FastSpeech2) to construct an end-to-end solution. This system can accurately translate Braille letters and sentences into speech, with measured recognition accuracy rates reaching 94.3% and 93.9%, respectively. This provides a more natural reading habit (sliding mode) and an efficient auxiliary technology solution for Braille reading that offers immediate auditory feedback. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the sensor structure in an embodiment of the present invention; Figure 2 This is a schematic diagram of the sensor manufacturing method in an embodiment of the present invention; Figure 3This is a scanning electron microscope image of the conductive thin film with a dual-gradient pore structure in an embodiment of the present invention; Figure 4 This is a hardware circuit diagram of the signal processing module in an embodiment of the present invention; Figure 5 This is a schematic diagram of a real-time Braille-to-speech recognition system in an embodiment of the present invention; Figure 6 This is a block diagram of the real-time Braille-to-speech recognition system in an embodiment of the present invention; Figure 7 This is a schematic diagram of a real-time Braille-to-speech recognition system used in an embodiment of the present invention.

[0024] Reference numerals: 1. Conductive thin film with dual gradient pore structure; 2. Polyimide thin film; 3. Interdigitated electrode; 4. Positioning frame. Detailed Implementation

[0025] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments. It should be noted that, without conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments.

[0026] Example 1 like Figure 1 As shown, an embodiment of the present invention provides a sensor, including: A conductive thin film with a dual gradient pore structure 1 is formed by electrospinning thermoplastic polyurethane elastomer and then treating it with ultrasonic multi-walled carbon nanotubes and polydimethylsiloxane solution. A polyimide film 2, wherein the polyimide film 2 has a double-layer structure, and the dual-gradient pore structure conductive film 1 is located in the double-layer structure of the polyimide film 2 to protect the dual-gradient pore structure conductive film 1. Interdigitated electrode 3 is in direct contact with the dual-gradient porous conductive film 1. The interdigitated electrode 3 is located in the double-layer structure of the polyimide film 2 to protect the interdigitated electrode 3.

[0027] The sensor in this embodiment of the invention is a flexible piezoresistive sensor with a dual-gradient pore structure. Furthermore, the sensor can be expanded into a 3*2 flexible piezoresistive sensor array using screen printing technology.

[0028] Specifically, the polyimide film 2 serves as a protective layer for the sensor, encapsulating the flexible piezoresistive sensor and providing waterproofing, dustproofing, and protection for the core conductive film. Furthermore, the polyimide film 2 is flexible enough to adhere to human skin, while its thickness provides strong support for the sensor.

[0029] Furthermore, the dual-gradient porous structure conductive film 1 is formed by adjusting the electrospinning process parameters.

[0030] like Figure 3 As shown, the dual-gradient porous conductive film 1 is a conductive film formed by electrospinning thermoplastic polyurethane elastomer and then treating it with ultrasonic multi-walled carbon nanotubes and polydimethylsiloxane solution. By controlling the setting of spinning process parameters, the film morphology exhibits a dual-gradient porous structure. Due to the relay amplification effect of the gradient porous structure, the sensor obtains a large linear range and sensitivity when subjected to external force. By incorporating ultrasonic multi-walled carbon nanotubes, a stable conductive network is formed on the surface of the spun fibers formed from thermoplastic polyurethane elastomer. Subsequently, surface treatment with polydimethylsiloxane solution allows for control electrode contact, thereby avoiding the side effects of preload on sensitivity and effectively reducing initial current and tunneling current, thus improving sensor performance.

[0031] The fabrication of the flexible interdigitated electrode 3 allows for control over the electrode surface roughness and width / spacing ratio, thereby improving sensor performance. A rougher electrode surface can generate greater contact changes between the active material and the electrode under pressure load, thus improving sensor sensitivity. Furthermore, increasing the electrode width / spacing ratio can reduce the initial contact between the electrode and the material, thereby reducing the initial current of the sensor. The fabricated flexible interdigitated electrode 3 has a width / spacing ratio of 2 / 1 and is mounted on one side of the conductive film.

[0032] Furthermore, the polyimide film 2 is provided with a rectangular positioning frame 4 for positioning the dual-gradient porous structure conductive film 1 and the interdigitated electrode 3.

[0033] Specifically, a rectangular positioning frame 4 is adhered to the polyimide film 2. The positioning frame 4 is used to position the dual-gradient porous structure conductive film 1 and the flexible interdigitated electrode 3, so that the dual-gradient porous structure conductive film 1 and the interdigitated electrode 3 are located in the middle of the polyimide film 2.

[0034] Furthermore, the dimensions of the dual-gradient porous conductive film 1 and the interdigitated electrode 3 are both 3mm × 3mm.

[0035] Specifically, in this application, the conductive film has a size of 3mm × 3mm and a thickness on the micrometer level. The interdigitated electrode 3 has a size of 3mm × 3mm, a linewidth of 200μm, and a line spacing of 100μm. The polyimide film 2 has a size of 10mm × 10mm.

[0036] Furthermore, it also includes an encapsulation structure for encapsulating the edges of the polyimide film 2.

[0037] Furthermore, the encapsulation structure is a high-temperature adhesive tape.

[0038] Specifically, after the polyimide film 2, the dual-gradient porous structure conductive film 1, and the interdigitated electrode 3 are naturally stacked, the encapsulation structure is used to encapsulate the edges of the polyimide film 2. The encapsulation structure can be made of high-temperature tape, or other suitable materials can be selected to replace the high-temperature tape; it is not limited to this.

[0039] Example 2 like Figure 2 As shown, this embodiment of the invention provides a method for manufacturing a sensor as described in Embodiment 1, comprising the following steps: (1) Preparation of polyimide film 2: At room temperature, 2.11 g of 4',4-diaminodiphenyl ether solute was dissolved in 30 g of N-methylpyrrolidone solvent and stirred for 30 minutes with a stirrer; 2.3 g of pyromellitic dianhydride was weighed and slowly added to the solution, and stirred at a constant temperature for 2 hours in a water bath at 30°C; the solution was dropped onto the silicon wafer using a dropper, and the solution was spread evenly on the silicon wafer using a two-step spin coating method. The first step was set to 500 rpm for 5 seconds, and the second step was set to 3000 rpm for 1 minute; the silicon wafer was placed on a heating table and baked at 150°C for 5 minutes, and then baked at 250°C for 1 hour to form a polyimide film 2 of about 2 μm.

[0040] (2) Fabrication of gradient pore structure TPU film: First, 3.2 g of TPU was dissolved in 10 mL of tetrahydrofuran solution and stirred at room temperature for 4 hours to obtain a TPU precursor solution. 3.6 g of PVA powder was added to 36.4 mL of deionized water and stirred at 90°C for 4 hours to obtain a PVA precursor solution. Then, the two precursor solutions were injected into syringes with a capacity of 10 ml. The syringes were placed in an electrospinning machine, and the parameters of the electrospinning machine were set as follows: the initial solution feed rate was 0.5 ml / h, the electrospinning voltage was set to 13 kV, the distance between the collector and the needle was set to 15 cm, and a 22 G needle was selected for the spinning needle. The electrospinning process lasted for 10 hours. Then, the solution feed rate was adjusted to 1 mL / h, a 20 G needle was selected, the voltage was increased to 17.5 kV, and electrospinning continued for 5 hours. A roller collector was used, with a layer of tin foil wrapped around the roller. The roller speed was set to 400 rpm. After spinning, the tin foil was removed from the roller, and the film was placed in a 70℃ constant temperature drying oven for 4 hours to cure. The spun film was then peeled off to obtain a TPU / PVA electrospun membrane. Finally, the cured membrane was immersed in deionized water for 30 minutes to remove PVA, resulting in a TPU nanofiber membrane with a gradient porous structure.

[0041] (3) Fabrication of the conductive film: First, 0.1 wt% of MWCNTs were dispersed in a mixed solution of 50 mL deionized water and 50 mL ethanol, and then sonicated for 30 minutes to obtain a MWCNTs suspension. The TPU membrane with a gradient porous structure prepared in the previous step was then immersed in the suspension for ultrasonic adsorption treatment. After 1 hour, it was removed and washed with deionized water to remove unadsorbed substances from the surface. Next, 1 wt% of uncured two-component PDMS (PDMS to curing agent mass ratio of 10:1) was dispersed in hexane. The prepared MWCNTs / TPU composite membrane was immersed in the PDMS / hexane solution for 30 minutes. Finally, it was dried in an oven at 80°C for 4 hours to obtain the MWCNTs / PDMS / TPU conductive film.

[0042] (4) Fabrication of flexible interdigitated electrodes 3: Photoresist is dropped into the center of the polyimide film 2 using a dropper. The photoresist is spread evenly on the film using a two-step spin coating method. The first step is set to 1500 revolutions per minute for 20 seconds, and the second step is set to 8000 revolutions per minute for 40 seconds. The silicon wafer with photoresist is placed under a beam of light passing through a mask with the pattern of interdigitated electrodes 3. After some of the photoresist is dissolved, it is sprayed onto the wafer surface at a low speed using potassium hydroxide to form a mist in a tank. After development, it is sprayed and cleaned with distilled water. The developed silicon wafer is treated with soft oxygen plasma. A 50 nm gold film is deposited on the polyimide film 2 using a magnetron sputtering system. The wafer is dried at 120°C for 20 minutes in a vacuum drying oven to remove the remaining photoresist. The polyimide film is peeled off from the silicon wafer to form a flexible electrode.

[0043] (5) Place the conductive film and the interdigitated electrode 3 between the polyimide film 2 to encapsulate the conductive film, the polyimide film 2 and the interdigitated electrode 3.

[0044] Example 3 like Figure 4-7 As shown, this embodiment of the invention provides a real-time Braille-to-speech recognition system, comprising: A sensing module, comprising a sensor as described in one of the objectives of the present invention, wherein the sensor module is used to acquire analog electrical signals generated when a finger slides or presses on Braille dots via the sensor array; A signal processing module is used to convert the analog electrical signals acquired by the sensing module into digital signals; A data transmission module is used to wirelessly transmit the digital signal obtained by the signal processing module to a mobile device via Bluetooth Low Energy. A character recognition module, configured in the mobile device, is used to input the digital signal transmitted by the data transmission module into a trained convolutional neural network, recognize and output the corresponding Braille character sequence; A speech synthesis module, configured in the mobile device, is used to input the Braille character sequence output by the character recognition module into a text-speech synthesis neural network to generate and output the corresponding speech signal.

[0045] The real-time Braille-to-speech recognition system of this invention can be applied to wearable electronic devices that support real-time Braille-to-speech recognition.

[0046] like Figure 4 As shown, the signal processing module is implemented using an ESP32-C6 microcontroller and an ADS1220-based prototype board. The MCU integrates Bluetooth Low Energy functionality, facilitating the development of mobile data acquisition systems. The sensor front end is connected via an onboard connector and powered by a fixed-voltage LDO regulator (3.3 V). Analog signals are sampled through sensing resistors, the actual resistance of which is determined by the surface resistance of the sensor and electrodes, as well as the contact resistance between them. The signal then passes through an anti-aliasing RC low-pass filter and is subsequently digitized by a Texas Instruments ADS1220 chip. The device supports 24-bit resolution and up to four analog input channels. Its internal ADC performs time-division multiplexing of the analog front end signal. The ADS1220's embedded programmable gain amplifier is configured with a gain of 1 to ensure maximum equivalent input impedance. The ADS1220 is further configured by the ESP32 to sample each channel at 24-bit resolution and 400 Hz. The digitized data is transmitted to the ESP32-C6 via SPI and then to the mobile device for further processing via BLE. The machine learning algorithm, deployed on a mobile device, comprises two networks: a data-to-text network and a text-to-speech network. The data-to-text network uses a CNN to process and classify Braille signals.

[0047] Furthermore, the convolutional neural network in the character recognition module includes 5 convolutional layers, 4 max-pooling layers, and 2 fully connected layers, with a convolutional kernel size of 4, a stride of 1, and a max-pooling layer size of 2.

[0048] Specifically, the network architecture consists of 5 convolutional layers, 4 max-pooling layers, and 2 fully connected layers. Activation functions are applied after each convolutional and fully connected layer to enhance the network's expressive power. The kernel size and stride of the convolutional layers are set to 4 and 1, respectively, while the max-pooling layer size is set to 2. The number of filters in each convolutional layer is 16, 32, 64, 128, and 128, respectively. During training, the learning rate is dynamically adjusted between 0.01 and 0.001, gradually decreasing. In the deployed framework, it can be replaced without any problems when a more suitable text-to-speech network becomes available.

[0049] Furthermore, the text-to-speech synthesis neural network in the speech synthesis module is a FastSpeech2 neural network, employing a transformer-based decoder.

[0050] In this embodiment, the system uses FastSpeech2, which employs an advanced transformer-based decoder, providing powerful text modeling capabilities and excellent robustness, designed to quickly generate high-quality, natural speech. Figure 5 The image shows a real-time Braille-to-speech conversion scenario using wearable electronic devices.

[0051] like Figure 5 As shown, a flexible piezoresistive sensor array is applied to a finger using a polyurethane film dressing for sliding recognition of Braille information. Continuous and distinguishable current peaks generated by the sensor array in response to the encoded Braille letters indicate successful recognition. The current data generated by the sensors is transmitted back to the terminal via a wireless module. The CNN network on the terminal can accurately recognize 26 Braille letters and one space symbol. Theoretically, it can accurately construct any sentence. The terminal determines the word structure by recognizing the space symbol and immediately sends the word to the text-to-speech network, then outputs the pronunciation of the word through a speaker. Notably, the number of sentences can be easily expanded. Furthermore, this embodiment simplifies the rules for encoding words and constructing sentences because only one Braille letter rule is applied. Experimental results show that this system is a convenient, intelligent, and reliable system for rapid and accurate Braille recognition, providing Braille readers with a more user-friendly electronic device interaction experience.

[0052] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A sensor, characterized in that, include: A conductive film with a dual gradient pore structure is formed by electrospinning thermoplastic polyurethane elastomer and then treating it with ultrasonic multi-walled carbon nanotubes and polydimethylsiloxane solution. A polyimide film, wherein the polyimide film has a double-layer structure, and the dual-gradient pore structure conductive film is located in the double-layer structure of the polyimide film to protect the dual-gradient pore structure conductive film. Interdigitated electrodes are in direct contact with the dual-gradient porous conductive film. The interdigitated electrodes are located in the bilayer structure of the polyimide film to protect them.

2. The sensor according to claim 1, characterized in that, The dual-gradient porous structure conductive film is formed by adjusting the electrospinning process parameters.

3. The sensor according to claim 1, characterized in that, The polyimide film has a rectangular positioning frame for positioning the dual-gradient porous conductive film and the interdigitated electrode.

4. The sensor according to claim 1, characterized in that, The dimensions of the dual-gradient porous conductive film and the interdigitated electrode are both 3mm × 3mm.

5. The sensor according to claim 1, characterized in that, It also includes an encapsulation structure for encapsulating the edges of the polyimide film.

6. The sensor according to claim 5, characterized in that, The encapsulation structure is a high-temperature adhesive tape.

7. A method for manufacturing a sensor as described in any one of claims 1-6, characterized in that, Includes the following steps: (1) Preparation of polyimide film: Dissolve 4',4-diaminodiphenyl ether in N-methylpyrrolidone, add pyromellitic dianhydride, spin-coat and bake to obtain polyimide film; (2) Fabrication of gradient porous TPU film: TPU and PVA are respectively made into precursor solutions, and after electrospinning, curing and PVA removal, a gradient porous TPU nanofiber film is obtained. (3) Fabrication of a conductive film with a dual gradient pore structure: The TPU nanofiber membrane is immersed in a MWCNTs suspension after ultrasonic treatment, and after cleaning, it is immersed in a hexane solution containing PDMS. After drying, a conductive film composed of MWCNTs, PDMS and TPU is obtained. (4) Fabrication of flexible interdigitated electrodes: Photoresist is spin-coated onto the polyimide film, and after exposure, development, plasma treatment, magnetron sputtering deposition of gold film, and removal of photoresist, flexible interdigitated electrodes are obtained; (5) The dual-gradient porous structure conductive film and the interdigitated electrode are placed in the polyimide film for encapsulation.

8. A real-time Braille-to-speech recognition system, characterized in that, include: A sensing module, comprising a sensor as described in any one of claims 1-6, wherein the sensor module is used to acquire analog electrical signals generated when a finger slides or presses on Braille dots via the sensor array; A signal processing module is used to convert the analog electrical signals acquired by the sensing module into digital signals; A data transmission module is used to wirelessly transmit the digital signal obtained by the signal processing module to a mobile device via Bluetooth Low Energy. A character recognition module, configured in the mobile device, is used to input the digital signal transmitted by the data transmission module into a trained convolutional neural network, recognize and output the corresponding Braille character sequence; A speech synthesis module, configured in the mobile device, is used to input the Braille character sequence output by the character recognition module into a text-speech synthesis neural network to generate and output the corresponding speech signal.

9. The identification system according to claim 8, characterized in that, The convolutional neural network in the character recognition module includes 5 convolutional layers, 4 max-pooling layers, and 2 fully connected layers. The convolutional kernel size is 4, the stride is 1, and the max-pooling layer size is 2.

10. The identification system according to claim 8, characterized in that, The text-to-speech synthesis neural network in the speech synthesis module is a FastSpeech2 neural network, which uses a transformer-based decoder.

Images