A deep learning-based full-cycle adaptive optical coating method

Abstract

The application belongs to the technical field of optical film and spectral intelligent sensing deep learning, and specifically discloses a full-cycle adaptive optical coating method based on deep learning, which comprises the following steps: collecting the light intensity and visible light spectrum distribution data of the external environment to obtain real-time light data; according to the data, identifying the daytime strong light or nighttime weak light scene through a light-weight convolutional neural network and generating a control instruction; converting the instruction into a driving voltage and outputting it to the conductive interlayer of a multi-layer film system; the multi-layer film system responds to the voltage to change the equivalent refractive index of the high refractive index film layer to adjust the optical path difference; under strong light, the optical path difference satisfies the deep destructive interference condition to reduce the reflectivity; under weak light, the destructive interference effect is weakened to reduce the round-trip scattering loss of light in the film layer, and the transmittance is improved. The application can make the optical coating of the telescope adaptively adjust the optical performance according to the change of light, so as to provide good observation effect under different light conditions all day round.

Description

Technical Field

[0001] This application belongs to the field of interdisciplinary technology of optical thin films and spectral intelligent sensing deep learning, and more specifically, relates to a deep learning-based full-cycle adaptive optical coating method. Background Technology

[0002] Telescopes are widely used in daily life and leisure activities such as outdoor sightseeing, astronomical observation, and science popularization experiences. They are a common optical device for observing natural phenomena. In actual use, strong daylight can easily cause overexposure of the image and deteriorate the observation effect, while insufficient ambient light at night can lead to a dim field of view and difficulty in clearly observing targets.

[0003] To address these issues, the industry commonly employs static FMC (Fluorescent Macroscopic Contrast) multilayer coating technology to optimize lens optical performance. This technology uses a fixed alternating structure of high and low refractive index films and applies coating to the entire optical interface, offering advantages such as a wide spectrum, high transmittance, and low reflectivity, resulting in superior imaging performance. However, static FMC multilayer coating only uses a fixed alternating high and low refractive index multilayer film system. Its film structure and optical parameters remain unchanged after fabrication, lacking ambient light perception and scene recognition capabilities. It is a passive optical structure, unable to automatically identify strong light and weak light scenes and switch operating modes. In strong daylight conditions, the fixed anti-reflective capability is insufficient to adequately suppress specular glare, easily leading to overexposure and reflective interference. In low-light conditions at night, the fixed transmittance results in insufficient light intake, making it difficult to clearly observe dim targets. A single coating scheme with fixed parameters cannot simultaneously meet the dual requirements of daytime overexposure prevention and high-definition nighttime observation.

[0004] Therefore, how to enable the optical coating of a telescope to adaptively adjust its optical performance according to changes in ambient light, so as to provide good observation and imaging effects under different lighting conditions in all weather conditions, is a problem that urgently needs to be solved. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the purpose of this application is to provide a deep learning-based full-cycle adaptive optical coating method that enables the optical coating of a telescope to adaptively adjust its optical performance according to changes in ambient light, thereby providing good observation and imaging effects under different lighting conditions in all weather conditions.

[0006] To achieve the above objectives, in a first aspect, this application provides a deep learning-based full-cycle adaptive optical coating method applied to an optical coating assembly. The optical coating assembly includes a multilayer film system comprising alternately stacked high-refractive-index and low-refractive-index layers, and a conductive interlayer embedded within the layers. The conductive interlayer receives an external driving voltage and introduces the driving voltage into the multilayer film system. The high-refractive-index layer is made of a material whose equivalent refractive index changes with the driving voltage under the influence of an electric field. The multilayer film system is used to change the equivalent refractive index of the high-refractive-index layer upon receiving the driving voltage to adjust optical performance, comprising the following steps: S10 collects data on the light intensity and visible light spectrum distribution of the external environment to obtain real-time light data; S20, using a pre-built lightweight convolutional neural network model to perform scene recognition on the real-time illumination data, and generating control commands based on the recognition results, wherein the recognition results are either daytime strong light scenes or nighttime weak light scenes; S30, the control command is converted into a corresponding driving voltage and output to the conductive electrode; The multilayer film system changes its equivalent refractive index in response to the driving voltage, thereby adjusting the optical path difference of the entire film system. In the daytime strong light scenario, the optical path difference is adjusted to meet the deep destructive interference condition to reduce the reflectivity of the lens surface. In the nighttime low light scenario, the optical path difference is adjusted to weaken the destructive interference effect to reduce the round-trip scattering loss of light inside the film layer and improve the overall transmittance of the lens.

[0007] As a further preferred embodiment, the high-refractive-index film layer is made of TiO2 material, and the low-refractive-index film layer is made of SiO2 material. The relationship between the equivalent refractive index and the driving voltage is determined by the following formula:

[0008] in, This represents the positive DC drive voltage applied to the conductive interlayer. Indicates the driving voltage is The equivalent refractive index of a single-layer TiO2 thin film, This represents the static intrinsic refractive index of TiO2 at 0V. Indicates the voltage linearity coefficient of the refractive index; The optical path difference of the entire film system is determined by the following formula:

[0009] in, This represents the total round-trip optical path difference of the entire film system. This indicates the physical thickness of a single TiO2 film. This represents the fixed refractive index of SiO2. This indicates the physical thickness of a single-layer SiO2 film.

[0010] As a further preferred embodiment, the relationship between the optical path difference and reflectivity is determined by the following formula:

[0011] in, This represents the total reflectivity of the lens surface as a function of the optical path difference. , This represents the Fresnel reflection coefficient at the interface of each medium layer. This represents the total phase difference between the reflected light at each interface of the multilayer film. This represents the phase delay term caused by the optical path difference in the reflected light from adjacent interfaces; The total phase difference Determined by the following formula:

[0012] in, Indicates the incident wavelength of visible light. This represents the fixed phase shift formed by the superposition of half-wave losses at each interface, and its value satisfies... 2 k ( (The parameter is a positive integer) and its range is 0 ≤ ≤8 ; The relationship between the optical path difference and the refractive index is determined by the following formula:

[0013] in, It represents the overall average transmittance of visible light from the lens. This indicates the total loss within the film layer; Total loss inside the membrane layer Determined by the following formula:

[0014] in, Indicates the total number of coating layers. Indicates the first Extinction coefficient of the thin film, Indicates the first The equivalent refractive index of the layered media is as follows: the equivalent refractive index of the SiO2 layer is a fixed value of 1.46, and the equivalent refractive index of the TiO2 layer is... , Indicates the first Thin film thickness, Indicates the wavelength of the incident visible light. Indicates the first The scattering loss coefficient of each medium interface, Show the first Fresnel reflection coefficient of each interface.

[0015] As a further preferred embodiment, in the multilayer film system, the high-refractive-index film layer is made of TiO2 material, with a single-layer thickness of 45 nm to 55 nm and a static refractive index of 2.42; the low-refractive-index film layer is made of SiO2 material, with a single-layer thickness of 85 nm to 95 nm and a static refractive index of 1.46; the total thickness of the entire film system is 680 ± 20 nm. The optical coating assembly comprises eight layers, which, from the substrate, are as follows: a first high-refractive-index film layer, a first low-refractive-index film layer, a second high-refractive-index film layer, the conductive interlayer, a third high-refractive-index film layer, a second low-refractive-index film layer, a fourth high-refractive-index film layer, and a third low-refractive-index film layer. The conductive interlayer is made of ITO material with a single-layer thickness of 12 nanometers. Both sides of the conductive interlayer are bonded to the high-refractive-index film layer, and there is no oxide isolation layer at the interface. Atomic-level interface bonding is achieved through synchronous sputtering co-deposition.

[0016] As a further preferred embodiment, the lightweight convolutional neural network model includes two 3×3 convolutional layers, a 2×2 max pooling layer, a ReLU activation layer, a global average pooling layer, and a binary classification output layer.

[0017] As a further preferred embodiment, the lightweight convolutional neural network model is pre-trained using a training dataset containing 120,000 sets of illumination samples, including 60,000 sets of daytime strong light scene samples and 60,000 sets of nighttime weak light scene samples. The daytime strong light scene samples cover four types of strong light conditions: sunny noon, cloudy daytime, strong light on snow, and reflection on water surfaces, as well as three types of shading conditions: partial tree shading, building shadows, and unshaded direct sunlight, with a light intensity range of 80,000 lux to 120,000 lux. The nighttime weak light scene samples cover four types of weak light conditions: twilight, urban streetlight environment, no moonlight outdoors, and low indoor light, with a light intensity range of 0.1 lux to 1,000 lux.

[0018] As a further preferred embodiment, the step of converting the control command into a corresponding driving voltage and outputting it to the conductive interlayer specifically includes: the lightweight convolutional neural network model outputting a recognition level of 0.8V or 3.3V; using a voltage comparator with a threshold of 2.0V for discrimination; when the recognition level is higher than 2.0V, it is determined to be a daytime strong light scene, triggering the boost circuit to output a 3.8V DC voltage; when the recognition level is lower than 2.0V, it is determined to be a nighttime weak light scene, triggering the voltage regulation circuit to output a 1.5V DC voltage.

[0019] As a further preferred embodiment, the method is executed cyclically with a period of 50ms, continuously collecting illumination data and performing real-time identification by the lightweight convolutional neural network model, and automatically switching the output driving voltage according to changes in ambient illumination.

[0020] As a further preferred embodiment, the acquisition of ambient light intensity and visible light spectral distribution data specifically includes: acquiring spectral data in the visible light band from 380nm to 780nm through a light acquisition module. The light acquisition module integrates a miniature light intensity sensor and a spectral sensor. The light intensity detection range of the light acquisition module is from 0.1 lux to 120,000 lux, the sampling period is 50ms, and a 128-dimensional spectral feature array is output in a single step.

[0021] Secondly, this application provides a deep learning-based full-cycle adaptive optical coating system for implementing the method described in any one of the above, comprising: The light acquisition module is used to collect data on ambient light intensity and visible light spectrum distribution to obtain real-time light data. The deep learning processing module has a built-in lightweight convolutional neural network model, which is used to receive the real-time illumination data transmitted by the illumination acquisition module, perform scene recognition on the real-time illumination data and output control commands. The recognition result is a daytime strong light scene or a nighttime weak light scene. The electronic control module is used to receive the control command and convert the control command into a corresponding driving voltage and output it to the optical coating component; The optical coating assembly includes a multilayer film system and a conductive interlayer embedded within the film layers. The multilayer film system comprises alternately stacked high-refractive-index film layers and low-refractive-index film layers. The conductive interlayer is used to receive the driving voltage and introduce the driving voltage into the multilayer film system. The high-refractive-index film layer is made of a material whose equivalent refractive index changes with the driving voltage under the influence of an electric field. The multilayer film system changes its equivalent refractive index in response to the driving voltage, thereby adjusting the optical path difference of the entire film system. Under strong light, the optical path difference satisfies the deep destructive interference condition to reduce reflectivity, significantly reduce external reflections of the mirror, and eliminate glare. Under weak light, destructive interference is weakened, reducing the round-trip scattering loss of light within the film layers and improving the overall transmission flux of the lens.

[0022] This application has the following effects: (1) This application enables the optical coating of the telescope to adaptively adjust its optical performance according to changes in ambient light, thereby providing good observation and imaging effects under different lighting conditions in all weather conditions. The optical coating component adapted to this application is set with a multilayer film system of alternating high and low refractive indices and an embedded conductive interlayer. The high refractive index TiO2 film layer can adjust the equivalent refractive index by changing the carrier concentration with the electric field. When executing the method, the first step is to collect ambient light intensity and spectral data in real time, the second step is to identify strong light / weak light scenes through a lightweight convolutional neural network and output control commands, and the third step is to convert the commands into corresponding voltages and send them to the conductive interlayer. The conductive interlayer conducts the voltage evenly to all TiO2 film layers. Under strong light scenes during the day, a high voltage of 3.8V is output, and the refractive index of TiO2 rises to 2.60. By maximizing the difference in refractive index between high and low media, the optical path difference of multiple HL units in the entire film system is matched with the deep destructive interference condition, which significantly reduces the specular reflectivity and eliminates glare from strong light. In low-light scenarios at night, a low voltage of 1.5V is output, and the refractive index of TiO2 drops to 2.53, reducing the difference in refractive index between high and low media, weakening the destructive effect of the film layers, reducing light scattering and absorption loss between film layers, and improving light transmission. Relying on the synergistic cooperation of deep learning intelligent recognition and electronically adjustable film system, the optical performance of the coating can be adaptively switched in real time according to the illumination, suppressing reflection during the day and ensuring light intake at night, so as to achieve high-quality observation and imaging in all weather conditions. (2) The film structure design of this application is reasonable. The ITO conductive interlayer sandwiches the TiO2 high-refractive-index layer on both sides, avoiding the electric field interruption by the insulating SiO2. The refractive index of all high-refractive-index film layers can be controlled synchronously. The film layers are evenly arranged with the conductive interlayer as the center. The alternating stacking of high and low refractive index layers can offset the thermal stress of the film. It is not easy to peel or crack in high and low temperature and humid environments, and the service life of the film is longer. At the same time, the superposition of multiple HL interference units realizes broadband anti-reflection. The phase cancellation effect can be stably achieved in the entire visible light band, and there is no problem of color reflection. (3) This application uses a lightweight convolutional neural network to complete the illumination scene recognition. The model has a small amount of computation and a fast response speed. It does not require large computing power hardware and is suitable for the miniaturized telescope equipment. The optical state of the coating can be switched by adjusting the driving voltage. There is no mechanical adjustment structure. The overall structure is simple and the energy consumption is low. It is compatible with the existing magnetron sputtering coating process and the mass production processing cost is lower. Attached Figure Description

[0023] Figure 1 This is a framework diagram of the deep learning-based full-cycle adaptive optical coating system provided in this application; Figure 2 This is a flowchart of the full-cycle adaptive optics coating system provided in this application; Figure 3 This is a structural diagram of the film layer of the optical coating component provided in the embodiments of this application; Figure 4This is a simulation verification curve of the applied voltage and equivalent refractive index of the single-layer TiO2 thin film provided in the embodiments of this application; Figure 5 This is a comparison chart of reflectance curves of various films in the full wavelength range of 380nm to 780nm provided in the embodiments of this application; Figure 6 These are images of a static FMC coated telescope observed in low-light conditions at night. Figure 7 These are observation images provided by this solution under low-light conditions at night; Figure 8 These are static FMC coated telescope images observed in strong daylight conditions. Figure 9 These are observation images provided by this scheme under strong daylight conditions. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0025] To address the issue that existing telescope optical coating technologies, particularly static FMC multilayer coatings with fixed parameters, cannot dynamically adjust their optical response according to ambient light, this application provides a deep learning-based full-cycle adaptive optical coating system. Through a coordinated approach of "ambient light acquisition—intelligent scene recognition—dynamic adjustment of coating parameters," all-weather adaptive control of the telescope's optical coating is achieved.

[0026] As shown in Figures 1 and 2, the deep learning-based full-cycle adaptive optical coating system provided in this application includes: The light acquisition module integrates a miniature light intensity sensor and a spectral sensor, which is fixed to the optical front end of the telescope. It is used to collect data on the light intensity and visible light spectrum distribution of the external environment in real time, and continuously transmit the collected raw light data to the deep learning processing module. The deep learning processing module incorporates a lightweight convolutional neural network (CNN) model, pre-trained on multi-scene datasets. Its input receives illumination data from the illumination acquisition module, and its output connects to the electronic control module. This module is used for feature extraction and classification of real-time illumination data, automatically identifying whether the current scene is a daytime strong light scene or a nighttime low light scene, and outputting corresponding control commands based on the identification results. The electronic control module is electrically connected to both the deep learning processing module and the optical coating assembly. It receives control commands from the deep learning processing module and outputs corresponding voltage / electronic control signals to the multilayer coating assembly to dynamically adjust the optical working state of the coating. The optical coating component, serving as the optical actuator of this application, employs a multilayer film structure with alternating high-refractive-index and low-refractive-index layers, deposited on the surface of a telescope lens substrate. TiO2 is preferably used as the high-refractive-index material, and SiO2 as the low-refractive-index material, deposited layer by layer via radio frequency sputtering. This multilayer film system possesses a broad-spectrum optical response, and the film layers are configured with an electrically controlled response structure, capable of receiving external electrical signals to dynamically change the equivalent refractive index and equivalent optical thickness, thereby altering the thin-film interference effect.

[0027] The deep learning-based full-cycle adaptive optical coating system provided in this application combines ambient light acquisition, intelligent scene recognition, and dynamic adjustment of coating optical parameters. The light acquisition module collects ambient light intensity and spectral distribution data in real time. Based on the light data, the deep learning processing module identifies whether the current scene is strong daylight or weak nightlight and sends corresponding control commands to the electronic control module. According to the control commands, the electronic control module outputs voltage signals to adjust the equivalent refractive index and optical thickness of the multilayer optical coating, realizing dynamic switching between anti-reflection under strong light and anti-reflection under weak light.

[0028] This application has the following beneficial effects: (1) In daytime strong light scenarios, the system automatically switches to strong anti-reflection mode. By enhancing the destructive interference effect of the multilayer film system, it maximizes the reduction of the reflectivity of the lens surface, effectively suppresses reflected glare, avoids overexposure of the image, and reduces the stimulation of strong light on the human eye, thus having an eye protection effect. (2) In low-light scenarios at night, the system automatically switches to high-transmittance mode. By adjusting the optical parameters of the multilayer film system, the overall transmittance is improved, the amount of light entering is increased, and the imaging brightness and contrast in low-light environments are significantly improved, solving the problem of traditional coatings not being visible in low light. (3) Relying on the light acquisition module and the deep learning processing module, the system can automatically identify changes in ambient light and switch optical working modes in real time without manual intervention. This breaks through the technical bottleneck of traditional fixed coating which can only compromise and adapt, and realizes all-weather adaptive work. (4) Based on the TiO2 / SiO2 high and low refractive index alternating multilayer film system, the coating prepared by the radio frequency sputtering process has good uniformity and strong adhesion; coupled with the lightweight CNN model, the computing power consumption is low, and the size and power consumption of the device are not significantly increased, which is suitable for the miniaturization and portability requirements of telescopes, especially children's telescopes. (5) The overall scheme has a simple structure and strong stability. It can be widely used in various visual optical equipment such as civilian telescopes and optical observation mirrors. It has a wide range of applications and high promotion value.

[0029] The following are specific implementation examples of this application: Example 1: Telescope Integrated Adaptive Optical Coating System Overall hardware assembly: The light acquisition module, lightweight CNN processing chip, electronic control driver board, and TiO2 / SiO2 electronically controlled multi-layer coated lenses are all integrated inside the handheld telescope barrel; the light intensity and spectrum sensors are fixed at the front frame of the objective lens, with no external cables. All modules are directly connected via PCB board signal lines, with unidirectional signal transmission: light acquisition module → CNN processing module → electronic control module → coated conductive electrode.

[0030] Light acquisition hardware parameters: The spectrum acquisition band covers the entire visible light range of 380nm to 780nm, the light intensity detection range is 0.1 lux (low light at night) to 120000 lux (strong light at noon), the sampling period is 50ms, and a 128-dimensional spectral feature array is output and transmitted to the CNN model in a single output.

[0031] Lightweight CNN network parameters: The total number of network parameters is only 12,000, suitable for local computation on small MCUs, without cloud requirements; Network structure: 2 layers of 3×3 convolutional layers + 2×2 max pooling layers + ReLU activation + global average pooling + binary classification output layer; The training dataset contains 120,000 lighting samples (60,000 sets of daytime strong light and 60,000 sets of nighttime weak light), with a single inference time ≤2ms. The model output has only two instruction levels: 3.8V for strong light mode and 1.5V for weak light mode.

[0032] CNN Model Training with Multi-Scene Illumination Data: This embodiment uses a lightweight CNN model training dataset containing 120,000 illumination samples, 60,000 for strong light and 60,000 for weak light, fully covering different weather conditions, time of day, season, and outdoor shading conditions. Strong light samples (60,000 samples): Includes four typical strong light conditions: sunny noon, cloudy daytime, strong light on snow, and water reflection; covers different solar intensities in spring, summer, autumn, and winter; also includes three shading conditions: partial tree shading, building shadows, and unshaded direct sunlight, with light intensity ranging from 80,000 lux to 120,000 lux. Weak light samples (60,000 samples): Includes four low light conditions: twilight, urban streetlight environment, no moonlight outdoors, and low indoor light; covers multiple low light periods: early morning, late night, and evening; distinguishes between environments with and without artificial lighting and environments with and without building shading, with light intensity ranging from 0.1 lux to 1000 lux.

[0033] Robustness of Lightweight CNN under Continuously Gradual Illumination: To verify the model's recognition stability under continuously gradually changing illumination conditions such as sunrise, sunset, and rapid transitions between light and dark in cloudy conditions, a dynamic illumination simulation platform was built to simulate linearly varying illumination from 0.1 lux to 120,000 lux at a rate of 500 lux / s. 30,000 sets of continuously varying samples were continuously collected for testing: Sunrise gradual illumination (500 lux → 10,000 lux, slow increase): average recognition accuracy 99.02%; Sunset gradual illumination (10,000 lux → 0.5 lux, slow decrease): average recognition accuracy 98.95%; Rapid transitions between light and dark in cloudy conditions (1000 lux → 120,0 ... (6000 lux periodic switching): average recognition accuracy 98.71%; overall recognition accuracy under gradual change conditions ≥98.7%, with no frequent misjudgments or jump output phenomena, excellent recognition robustness in the face of continuous changes in brightness in natural environment, and stable output level command without jitter.

[0034] The structure of the film layer in optical coating components (e.g.) Figure 3 As shown): In this embodiment, the optical coating assembly has a total of 8 layers. The complete stacking order of the film layers (substrate → air) is: H1 (TiO2) → L1 (SiO2) → H2 (TiO2) → ITO → H3 (TiO2) → L2 (SiO2) → H4 (TiO2) → L3 (SiO2). The substrate is K9 optical glass. An ultra-thin ITO electrically controlled conductive layer is embedded in the interlayer of the film layers to form an electrically modulated structure. The high-refractive-index TiO2 single layer has a thickness of 45-55 nm and a static refractive index n=2.42; the low-refractive-index SiO2 single layer has a thickness of 85-95 nm and a static refractive index n=1.46. The total thickness of the entire film system is 680±20 nm. The ITO single layer has a thickness of 12 nm. The film layers are tightly bonded to TiO2 on both sides, and there is no oxide isolation layer at the interface. Atomic-level interface bonding is achieved by simultaneous sputtering co-deposition. The conductive layer extends laterally to the edge of the lens to lead out the positive and negative electrodes.

[0035] The significance of the layer arrangement design is as follows: Bottom layer H1-L1: TiO2 provides strong adhesion, and HL pairing eliminates internal reflections from the glass substrate. H2 sandwiches the left side of ITO: isolating and insulating SiO2, ensuring electric field energy conduction to all TiO2 on the left side. ITO sandwiched between H2 and H3: double-sided bonding of semiconductor TiO2, voltage synchronous regulation of all high-refractive-index layers, and stable electronically controlled destructive phase function. Right side H3-L2 and H4-L3 two sets of HL units: multi-layer destructive superposition significantly reduces external strong light reflection across the entire wavelength range. Top layer L3 (SiO2): reduces the refractive index difference with air, decreasing surface native reflection, while also providing wear-resistant protection for the inner film. Overall alternating high and low structure: film layer stresses cancel each other out; high pressure during the day reduces reflection through deep destructive phase reduction, while low pressure at night weakens reflection and increases light transmission.

[0036] Dedicated complete fabrication process for RF sputtering electro-controlled films: All film layers are deposited using magnetron RF sputtering in a single process, with optimized process parameters for electro-optically tunable TiO2 thin films. Base vacuum background: 8 × 10⁻⁶ -4 Pa; Sputtering power: TiO2 target 120W, SiO2 target 100W, ITO composite target 80W; Working gas: high-purity argon, sputtering pressure 0.3Pa; TiO2 film oxygen partial pressure control: a small amount of oxygen is introduced, the oxygen-argon partial pressure ratio is 1:12, to suppress excess oxygen vacancies and ensure linearly adjustable carrier voltage; In-situ annealing process: after each film deposition, vacuum annealing is carried out at 120℃ for 15min to eliminate internal stress and improve the conductivity and electro-optic response linearity of the ITO / TiO2 interface.

[0037] Compared to the simple sputtering process of existing static FMC multilayer coating, which only alternately deposits TiO2 and SiO2 dielectric films, the electrically controllable TiO2 / SiO2 / ITO composite multilayer film in this embodiment has the following four major difficulties in preparation, which are overcome one by one by this embodiment through a dedicated RF sputtering process: (1) The problem of interface matching between ITO conductive interlayer and TiO2 dielectric layer Static FMC has no built-in conductive layer, thus avoiding the problems of heterogeneous film lattice mismatch and excessively high interface impedance. In contrast, the ultrathin ITO interlayer in this design is mismatched with the TiO2 lattice and thermal expansion coefficient, easily leading to interface microcracks, ion diffusion, and increased conductivity impedance, resulting in electro-optic modulation failure. This embodiment employs continuous co-sputtering of TiO2 and ITO, coupled with vacuum low-temperature annealing at 120℃ after each layer deposition. This releases interface stress, reduces contact impedance, and simultaneously controls the ITO thickness to 12nm to suppress indium ion diffusion, ensuring uniform electric field control of the overall film refractive index.

[0038] (2) The oxygen vacancy concentration of TiO2 thin films is difficult to control precisely. Static FMC uses only pure argon sputtering without oxygen supply. Excessive oxygen vacancies in TiO2 lead to carrier saturation, resulting in a very small adjustable refractive index range and making it impossible to achieve dual-mode switching between strong and weak light. This embodiment uses a mixed sputtering atmosphere with an oxygen-argon partial pressure ratio of 1:12 to precisely constrain the intrinsic carrier concentration, leaving sufficient space for electric field adjustment, achieving linear and continuous refractive index variation in the 0~4V range; coupled with closed-loop gas control using a gas mass flow meter, it ensures the consistency of optical parameters in batch products.

[0039] (3) Thermal stress and process timing problems caused by layer-by-layer vacuum annealing Static FMC can continuously deposit all film layers in one go without an annealing process; in this embodiment, each film layer requires 15 minutes of vacuum annealing after deposition. Frequent temperature rises and falls can easily cause film layer cracking and peeling, and significantly increase the difficulty of equipment timing control. This embodiment designs a dedicated automated segmented sputtering program, adopts step-by-step slow temperature rises and falls, and high-vacuum annealing throughout the process to eliminate water vapor impurities. The film layer adhesion is stable at ≥5B, realizing fully automated closed-loop operation of deposition and annealing.

[0040] (4) The difficulty of precise thickness control in three-target synergistic sputtering increases Static FMC uses only TiO2 and SiO2 dual targets, making parameter matching simple. This embodiment adds an ITO composite target. The three target materials have significantly different sputtering power and deposition rates, which can easily lead to film thickness deviations from the design value, compromising the thin film interference optical performance. This embodiment pre-calibrates a three-target sputtering rate database, coupled with an independent regulated RF power supply and in-situ spectral thickness monitoring, to precisely control the nanometer-level thicknesses of TiO2, SiO2, and ITO. The total film thickness is stably maintained at 680±20nm, ensuring stable achievement of optical effects such as deep destructive interference under strong light and high transmission under weak light.

[0041] Electronic control module drive circuit design: The complete circuit link of the electronic control module is as follows: MCU outputs 0.8V / 3.3V to identify the level → voltage comparator distinguishes high and low levels → PWM boost chip + linear regulator circuit → outputs a stable 1.5V / 3.8V DC to the ITO conductive electrode. Level discrimination unit: threshold 2.0V; above 2.0V is determined to be strong light, triggering the 3.8V boost circuit; below 2.0V is determined to be weak light, triggering the 1.5V regulator circuit. Boost conversion circuit: input 3.3V onboard power supply, maximum output current 12mA, matching the ITO electrode load of the lens. Output protection: series 1kΩ current-limiting resistor to prevent short circuit breakdown of the film electrode. The lightweight CNN microcontroller MCU can only output low-voltage logic levels of 0.8V and 3.3V, which is insufficient to drive the TiO2 / ITO electro-optic film to achieve refractive index regulation. This circuit is used to distinguish and convert the high and low recognition levels output by the MCU into the 1.5V / 3.8V driving voltage required by the film layer, matching the electro-optical control requirements of TiO2, while realizing anti-interference, low power consumption and short circuit protection, and completing the hardware signal link from recognition to film coating optical adjustment.

[0042] Physical mechanism of electrically controlled refractive index change in TiO2 / SiO2 multilayer films: TiO2 is a wide bandgap semiconductor dielectric. Under the action of an applied positive electric field, the concentration of free carriers inside the film layer changes linearly with the applied voltage. According to the effective dielectric approximation (EMA) model, the change in carrier concentration directly changes the equivalent refractive index of TiO2, thereby changing the optical path difference and thin film interference effect of the entire film system. The specific control rules are as follows: Applying a 3.8V strong light driving voltage: The carrier concentration inside TiO2 increases, the overall equivalent refractive index of the film system increases, the optical path difference between the high and low film layers meets the deep destructive interference condition, significantly suppressing the reflection of the lens surface and achieving strong light anti-glare; Applying a 1.5V weak light driving voltage: The carrier concentration inside TiO2 decreases, the overall equivalent refractive index of the film system decreases, the deep destructive interference effect is weakened, the visible light spectrum energy loss is reduced, the overall transmission flux of the lens is increased, and it is suitable for low light observation scenarios at night.

[0043] In this embodiment, the relationship between the equivalent refractive index of TiO2 and the driving voltage is determined by formula (1): (1) In the formula, The positive DC drive voltage applied to the ITO conductive interlayer is 1.5V and 3.8V in this embodiment. is the equivalent refractive index of a single-layer TiO2 film at voltage V; The static intrinsic refractive index of TiO2 at 0V is a fixed value of 2.42. The refractive index voltage linearity coefficient is 0.065.

[0044] The total round-trip optical path difference of the entire film system is determined by formula (2): (2) In the formula, This represents the total round-trip optical path difference of the entire membrane system. The refractive index of TiO2 at the corresponding voltage is calculated by equation (1); The physical thickness of a single-layer TiO2 film is 45–55 nm. The refractive index of SiO2 is fixed at 1.46. The physical thickness of a single-layer SiO2 film is 85–95 nm.

[0045] The relationship between optical path difference and phase difference is determined by formula (3): (3) In the formula, This represents the total phase difference between the reflected light at each interface of the multilayer film. The incident wavelength of visible light; The fixed phase offset is formed by the superposition of half-wave losses at each interface, and its value satisfies the following conditions. 2 k ( (The parameter is a positive integer) and its range is 0 ≤ ≤8 .

[0046] The relationship between optical path difference and reflectivity is determined by formula (4): (4) In the formula, The total reflectivity of the lens surface varies with the optical path difference; , Let be the Fresnel reflection coefficient at the interface of each medium layer; This is the phase delay term caused by the optical path difference in the reflected light from adjacent interfaces. The total phase difference between the reflected light at each interface of the multilayer film is calculated by equation (3).

[0047] Total loss inside the film layer Determined by formula (5): (5) In the formula, This represents the total number of coating layers. For the first Extinction coefficient of the thin film; For the first The equivalent refractive index of the dielectric layer is given by equation (1). The equivalent refractive index of the SiO2 layer is a fixed value of 1.46. The equivalent refractive index of the TiO2 layer is calculated by equation (1) based on the corresponding voltage of the TiO2 refractive index. ; For the first Thin film thickness; The wavelength of the incident visible light; For the first Scattering loss coefficient of each medium interface; For the first Fresnel reflection coefficient of each interface.

[0048] The relationship between optical path difference and refractive index is determined by formula (6): (6) In the formula, The overall average transmittance of visible light of the lens The total internal loss of the membrane layer is calculated using equation (5); The total reflectance of the lens surface is calculated by equation (4).

[0049] Figure 4 Simulation verification curves of applied voltage versus equivalent refractive index for TiO2 thin films are presented. Figure 4 As can be seen, the horizontal axis of the curve represents the applied DC driving voltage of the ITO conductive layer, and the vertical axis represents the equivalent refractive index of the TiO2 film. It intuitively reflects the change law of TiO2 refractive index with applied voltage: the static refractive index of TiO2 is 2.42 when there is no applied voltage. As the driving voltage increases, the equivalent refractive index of TiO2 continues to rise and the growth rate gradually slows down, showing a saturation growth trend. The refractive index is 2.53 under the 1.5V nighttime working condition, and the refractive index reaches the saturation value of 2.60 under the 3.8V daytime strong light working condition. At the low voltage of 1.5V, the difference between the refractive indices of TiO2 and SiO2 is small, which weakens the destructive interference, reduces the light energy loss in the film, and increases the light transmittance of the lens, making it suitable for the low light environment at night. At the high voltage of 3.8V, the difference between the refractive indices of the high and low media is maximized. All groups of HL interference units in the entire film system synchronously meet the deep destructive interference condition, which greatly reduces the reflection of strong external light and suppresses glare. This verifies the feasibility of electric field-controlled refractive index and day and night adaptive optical performance of this structure.

[0050] End-to-end response latency breakdown: The system's latency from light acquisition to the stable switching of film optical parameters is divided into four segments: sensor acquisition latency 50ms; CNN inference computation latency ≤2ms; electronic control signal amplification and output latency ≤30ms; and film refractive index stabilization response latency ≤1118ms. The total latency for switching between strong light mode and weak light mode is ≤1.2s, and the total latency for switching between weak light mode is ≤1.0s. The system can quickly adapt and adjust its optical state after sudden changes in light intensity.

[0051] Closed-loop working logic: The telescope continuously cycles after powering on, acquiring light data every 50ms → CNN real-time recognition → outputting corresponding voltage signals to adjust the optical parameters of the film layer. No manual operation is required throughout the process, and the mode is automatically switched in real time when the ambient light changes abruptly.

[0052] Example 2: Alternative Solution for Modular Split External Controller Hardware deployment structure: The telescope tube retains only the light acquisition module and optical coating components; the lightweight CNN processing module and electronic control module are integrated into an independent external controller, which transmits data via 2.4G wireless Bluetooth communication with a communication latency of ≤80ms; the external controller can be continuously powered by an external 5V DC power supply.

[0053] Workflow differences: The front-end sensor collects spectral and light intensity data and wirelessly sends it to the external controller. After the CNN in the controller completes the scene determination, it sends the voltage adjustment command back to the telescope coating electrode to realize remote switching between day and night modes.

[0054] Applicable scenarios and limitations: It is suitable for large fixed astronomical telescopes, which can expand computing power and reduce heat generation inside the telescope tube; however, the overall size of the device is large and there is a slight delay in wired / wireless transmission, making it unsuitable for small handheld telescopes for children.

[0055] Gradient multi-illuminance comparison experiment data: The unified test conditions were the entire visible light band from 380 to 780 nm. The spectrophotometer simultaneously measured the average reflectance, average transmittance, and relative contrast of the image. Five typical illuminance gradients were selected to compare traditional single-layer coating, fixed FMC multilayer coating, and the adaptive coating of this embodiment. The results are shown in Table 1. Figure 5 A comparison of reflectance curves for various films across the entire wavelength range of 380nm to 780nm is presented.

[0056] Table 1. Multi-illuminance comparison experiment

[0057] Experimental Conclusion 1: Within the full illuminance range, conventional fixed FMC multilayer coating parameters are not adjustable, and can only achieve a compromise balance, failing to address both strong light glare and insufficient light intake in weak light. The system in this embodiment can dynamically output the corresponding driving voltage according to the real-time illuminance: automatically reducing reflectivity to 0.8%~0.9% in high illuminance to eliminate specular glare; automatically increasing overall transmittance to over 96.7% in low illuminance to significantly improve the brightness of low-light images; and automatically matching intermediate parameters during cloudy transition lighting, outperforming existing fixed coating products across the entire illuminance range.

[0058] Experimental Conclusion 2: The static FMC multilayer coating has a stable reflectance in the range of 3.0%~3.9%, and its spectral curve is fixed with no adjustment capability. It can only be adapted to a single lighting environment. Under strong daylight, the reflectance is too high and glare is easy to occur. Under weak nightlight, the light energy loss is large and the light intake is insufficient. It cannot simultaneously meet the needs of both scenarios. The electro-controlled coating in this embodiment has dual-mode adaptive control capability: in the strong light mode, the full-band reflectance is only 0.8%~0.9%, and the reflection of the entire visible light band is greatly suppressed, effectively eliminating the glare of the mirror surface. In the weak light mode, the full-band reflectance is maintained at 3.0%~3.1%, and the reflection loss is significantly lower than that of the static FMC coating. At the same time, the transmitted light is enhanced, which fully improves the light intake and screen brightness in the weak light environment. In the transitional illumination environment such as cloudy and dawn / dusk, the film can output an intermediate refractive index state to achieve a smooth transition of optical performance. The optical performance is better than that of the static FMC multilayer coating.

[0059] Comparison of observation images from a traditional static FMC coated telescope in low-light conditions at night ( Figure 6 ) and the observation images of this scheme ( Figure 7 As can be seen, static FMC coating has low transmittance, ranging from 91.3% to 91.6%, resulting in poor imaging contrast. Real-world images are generally dark, with many faint stars lost, and severe background stray light and haze, failing to capture the details of the Milky Way. This solution significantly improves visible light transmittance to over 96.7% in low-light mode, resulting in a significant improvement in imaging contrast. Real-world nighttime shooting can clearly capture a large number of stars and Milky Way nebulae, with image brightness, purity, and detail levels all superior to traditional static FMC coating.

[0060] Comparison of observation images from a traditional static FMC coated telescope under strong daylight conditions ( Figure 8 ) and the observation images of this scheme ( Figure 9 As can be seen, the static FMC coating has a reflectivity as high as 3.0%~3.9%, resulting in severe stray light from strong light on the lens surface. In actual shooting, the sky highlights appear white, the scenery appears gray, the water surface has a strong foggy reflection, the image contrast is low, and the highlight details are easily lost. In this embodiment, the reflectivity of the film layer is reduced to 0.8%~0.9% under strong light mode, effectively suppressing specular glare in various environments. The sky in the actual shooting image has pure color, the colors of buildings and vegetation are accurately reproduced, the stray light interference from the water surface is eliminated, the image contrast is significantly improved, and the image transparency and detail performance under strong light are comprehensively superior to the static FMC coating.

[0061] The system architecture of this application can be implemented in the following two alternative schemes: Category 1: Adaptive solutions for single-layer coatings The overall control system of this solution (light acquisition module, deep learning processing module, and electronic control module) is completely consistent with the main solution, except that the alternating high and low refractive index multilayer film system is replaced with a single-layer optical coating. The single-layer coating uses an electronically adjustable optical thin film material. During operation, after the deep learning module identifies strong light / weak light scenes, it changes the equivalent refractive index and film thickness of the single-layer coating through electronic control signals, switching between anti-reflection and anti-reflection working states to achieve basic day and night dual-effect functionality.

[0062] This solution has a simple structure and lower coating preparation cost, but it is limited by the inherent characteristics of narrow spectral response of single-layer coating. Its anti-reflection and anti-transmission effects across the entire wavelength range are limited, and its strong light glare suppression and weak light brightness enhancement effects are weaker than those of multi-layer film solutions. It is only suitable for economical optical equipment with low requirements for imaging quality.

[0063] Category 2: Modular Deployment Solution The optical coating assembly of this solution still adopts the core high and low refractive index alternating multilayer film system of this embodiment; the deep learning processing module and the electronic control module are separated from the telescope body and adopted as an external independent controller, which communicates with the light acquisition module and coating assembly through wired / wireless means. During operation, the light acquisition module collects data and transmits it to the external deep learning controller to complete scene recognition and command output, and then remotely controls the telescope coating to switch working modes.

[0064] This approach reduces the computational power and hardware load on the telescope itself, making it suitable for large, fixed observation equipment.

[0065] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A deep learning-based full-cycle adaptive optical coating method, characterized in that, This invention relates to an optical coating assembly, comprising a multilayer film system. The multilayer film system includes alternately stacked high-refractive-index and low-refractive-index layers, and a conductive interlayer embedded within the layers. The conductive interlayer receives an external driving voltage and introduces the driving voltage into the multilayer film system. The high-refractive-index layers are made of a material whose equivalent refractive index changes with the driving voltage under the influence of an electric field. The multilayer film system is used to adjust the optical performance by changing the equivalent refractive index of the high-refractive-index layers when receiving the driving voltage, comprising the following steps: S10 collects data on the light intensity and visible light spectrum distribution of the external environment to obtain real-time light data; S20, using a pre-built lightweight convolutional neural network model to perform scene recognition on the real-time illumination data, and generating control commands based on the recognition results, wherein the recognition results are either daytime strong light scenes or nighttime weak light scenes; S30, the control command is converted into a corresponding driving voltage and output to the conductive electrode; The multilayer film system changes its equivalent refractive index in response to the driving voltage, thereby adjusting the optical path difference of the entire film system. In the daytime strong light scenario, the optical path difference is adjusted to a deep destructive interference condition to reduce the reflectivity of the lens surface, significantly reduce external reflection of the mirror, and eliminate glare. In the nighttime low light scenario, the destructive interference is weakened to reduce the round-trip scattering loss of light inside the film layer and improve the overall transmission flux of the lens.

2. The deep learning-based full-cycle adaptive optical coating method as described in claim 1, characterized in that, The high-refractive-index film layer is made of TiO2 material, and the low-refractive-index film layer is made of SiO2 material. The relationship between the equivalent refractive index and the driving voltage is determined by the following formula: in, This represents the positive DC drive voltage applied to the conductive interlayer. Indicates the driving voltage is The equivalent refractive index of a single-layer TiO2 thin film, This represents the static intrinsic refractive index of TiO2 at 0V. Indicates the voltage linearity coefficient of the refractive index; The optical path difference of the entire film system is determined by the following formula: in, This represents the total round-trip optical path difference of the entire film system. This indicates the physical thickness of a single TiO2 film. This represents the fixed refractive index of SiO2. This indicates the physical thickness of a single-layer SiO2 film.

3. The deep learning-based full-cycle adaptive optical coating method as described in claim 1 or 2, characterized in that, The relationship between the optical path difference and reflectivity is determined by the following formula: in, This represents the total reflectivity of the lens surface as a function of the optical path difference. , …represents the Fresnel reflection coefficient at the interface of each medium layer. This represents the total phase difference between the reflected light at each interface of the multilayer film. This represents the phase delay term caused by the optical path difference in the reflected light from adjacent interfaces; The total phase difference Determined by the following formula: in, Indicates the incident wavelength of visible light. This represents the fixed phase shift formed by the superposition of half-wave losses at each interface, and its value satisfies... 2 k The parameter range is 0≤ ≤8 ; The relationship between the optical path difference and the refractive index is determined by the following formula: in, It represents the overall average transmittance of visible light from the lens. This indicates the total loss within the film layer; Total loss inside the membrane layer Determined by the following formula: in, Indicates the total number of coating layers. Indicates the first Extinction coefficient of the thin film, Indicates the first The equivalent refractive index of the layered medium, the equivalent refractive index of the SiO2 layer is a fixed value, and the equivalent refractive index of the TiO2 layer is [missing value]. , Indicates the first Thin film thickness, Indicates the wavelength of the incident visible light. Indicates the first The scattering loss coefficient of each medium interface, Show the first Fresnel reflection coefficient of each interface.

4. The deep learning-based full-cycle adaptive optical coating method as described in claim 1, characterized in that, In the multilayer film system, the high refractive index film layer is made of TiO2 material, with a single layer thickness of 45 nm to 55 nm and a static refractive index of 2.42; the low refractive index film layer is made of SiO2 material, with a single layer thickness of 85 nm to 95 nm and a static refractive index of 1.46; the total thickness of the entire film system is 680 ± 20 nm. The optical coating assembly comprises eight layers, which, from the substrate, are as follows: a first high-refractive-index film layer, a first low-refractive-index film layer, a second high-refractive-index film layer, the conductive interlayer, a third high-refractive-index film layer, a second low-refractive-index film layer, a fourth high-refractive-index film layer, and a third low-refractive-index film layer. The conductive interlayer is made of ITO material with a single-layer thickness of 12 nanometers. Both sides of the conductive interlayer are bonded to the high-refractive-index film layer, and there is no oxide isolation layer at the interface. Atomic-level interface bonding is achieved through synchronous sputtering co-deposition.

5. The deep learning-based full-cycle adaptive optical coating method as described in claim 1, characterized in that, The lightweight convolutional neural network model includes two 3×3 convolutional layers, a 2×2 max pooling layer, a ReLU activation layer, a global average pooling layer, and a binary classification output layer.

6. The deep learning-based full-cycle adaptive optical coating method as described in claim 5, characterized in that, The lightweight convolutional neural network model was pre-trained using a training dataset containing 120,000 light samples, including 60,000 daytime strong light scene samples and 60,000 nighttime weak light scene samples. The daytime strong light scene samples cover four types of strong light conditions: sunny noon, cloudy daytime, strong light on snow, and water reflection, as well as three types of shading conditions: partial tree shading, building shadows, and unshaded direct sunlight, with light intensity ranging from 80,000 lux to 120,000 lux. The nighttime weak light scene samples cover four types of weak light conditions: twilight, urban streetlight environment, no moonlight outdoors, and low indoor light, with light intensity ranging from 0.1 lux to 1,000 lux.

7. The deep learning-based full-cycle adaptive optical coating method as described in claim 1, characterized in that, The step of converting the control command into a corresponding driving voltage and outputting it to the conductive interlayer specifically includes: the lightweight convolutional neural network model outputting a recognition level of 0.8V or 3.3V; using a voltage comparator with a threshold of 2.0V for discrimination; when the recognition level is higher than 2.0V, it is determined to be a daytime strong light scene, triggering the boost circuit to output a 3.8V DC voltage; when the recognition level is lower than 2.0V, it is determined to be a nighttime weak light scene, triggering the voltage regulation circuit to output a 1.5V DC voltage.

8. The deep learning-based full-cycle adaptive optical coating method as described in claim 7, characterized in that, The method is executed cyclically with a period of 50ms, continuously collecting illumination data and performing real-time identification by the lightweight convolutional neural network model, and automatically switching the output driving voltage according to changes in ambient illumination.

9. The deep learning-based full-cycle adaptive optical coating method as described in claim 1, characterized in that, The acquisition of ambient light intensity and visible light spectral distribution data specifically includes: acquiring spectral data in the visible light band from 380nm to 780nm through a light acquisition module. The light acquisition module integrates a miniature light intensity sensor and a spectral sensor. The light intensity detection range of the light acquisition module is from 0.1 lux to 120,000 lux, the sampling period is 50ms, and a 128-dimensional spectral feature array is output in a single output.

10. A deep learning-based full-cycle adaptive optical coating system, characterized in that, To implement the method of any one of claims 1 to 9, comprising: The light acquisition module is used to collect data on ambient light intensity and visible light spectrum distribution to obtain real-time light data. The deep learning processing module has a built-in lightweight convolutional neural network model, which is used to receive the real-time illumination data transmitted by the illumination acquisition module, perform scene recognition on the real-time illumination data and output control commands. The recognition result is a daytime strong light scene or a nighttime weak light scene. The electronic control module is used to receive the control command and convert the control command into a corresponding driving voltage and output it to the optical coating component; The optical coating assembly includes a multilayer film system and a conductive interlayer embedded within the film layers. The multilayer film system comprises alternately stacked high-refractive-index film layers and low-refractive-index film layers. The conductive interlayer is used to receive the driving voltage and introduce the driving voltage into the multilayer film system. The high-refractive-index film layer is made of a material whose equivalent refractive index changes with the driving voltage under the influence of an electric field. The multilayer film system changes its equivalent refractive index in response to the driving voltage, thereby adjusting the optical path difference of the entire film system. Under strong light, the optical path difference satisfies the deep destructive interference condition to reduce reflectivity, significantly reduce external reflections of the mirror, and eliminate glare. Under weak light, destructive interference is weakened, reducing the round-trip scattering loss of light within the film layers and improving the overall transmission flux of the lens.

Images