Self-adaptive vehicle visual inspection system based on deep learning
By performing multiple optimization processes on the image signal through the signal preprocessing module of the adaptive vehicle vision inspection system, the problem of noise interference in complex environments is solved, and the image quality and detection accuracy of the deep learning model are improved, making it suitable for intelligent transportation and autonomous driving fields.
Patent Information
- Application Number
- CN202410982871.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-22
- Publication Date
- 2024-11-08
AI Technical Summary
Existing vehicle vision inspection systems are affected by noise sources such as electromagnetic interference, signal amplification drift, and power frequency noise in complex environments, which leads to a decrease in image quality and consequently affects the detection accuracy of deep learning models.
An adaptive vehicle vision inspection system is adopted, including a CCD image sensor, a signal preprocessing module, a data processing module, and a user interface. The signal preprocessing module performs multiple optimization processes on the image signal through a pre-amplification and noise reduction unit, a power frequency notch depth conditioning unit, and an opto-isolation noise reduction unit to eliminate noise interference.
It significantly improves the purity and stability of image signals, enhances the adaptability and detection accuracy of deep learning models in complex environments, and achieves higher precision vehicle visual detection.
Smart Images

Figure CN118921548A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of vehicle vision detection, and in particular to an adaptive vehicle vision detection system based on deep learning. Background Art
[0002] In recent years, the rapid development of deep learning technology has brought new breakthroughs in vehicle visual inspection. Deep learning models, especially convolutional neural networks (CNNs), can automatically learn and extract high-level abstract features from large amounts of data. These features have stronger expressiveness and robustness for vehicle detection and recognition. However, the efficient operation of deep learning models depends on high-quality and high-purity image data as input. In practical applications, vehicle images are often interfered by a variety of noise sources, including sensor noise, circuit noise, environmental noise, etc. These noises will seriously affect the quality of the image, thereby reducing the detection performance of the deep learning model.
[0003] At present, most image acquisition systems use a CMOS image sensor combined with an FPGA field programmable logic gate array to collect and process image data. Referring to the existing patent document (publication number CN 220554076U), the utility model patent discloses a CCD data high-speed acquisition device based on FPGA, which outputs the optical signal to be detected sensed by the CCD image sensor to an analog-to-digital conversion chip. Under the action of the clock and integration time signal provided by the FPGA chip, the analog-to-digital conversion chip converts the analog signal output by the CCD image sensor into a digital signal and stores it in the buffer of the FPGA chip. The FPGA chip transmits the data in the data buffer to the host computer through the USB interface transmission chip. This technical solution is susceptible to noise interference during the conversion of analog signals into digital signals and the transmission process, which will reduce the purity of the image signal, thereby affecting the accuracy of subsequent image processing.
[0004] The existing invention patent document (publication number CN 108681270 A) discloses a signal processing circuit of a linear array CCD detector, including a preamplifier circuit, a differential amplifier circuit, a filter circuit, an analog-to-digital converter, a light intensity adjustment circuit and a light source connected in sequence; wherein the preamplifier circuit is used to amplify the current of the signal output by the linear array CCD detector to improve the load-carrying capacity; the differential amplifier circuit is used to differentially amplify the output signal of the preamplifier circuit to suppress the interference caused by the common mode and obtain an effective differential mode signal; the filter circuit is used to filter out the high-frequency components in the differential mode signal output by the differential amplifier circuit. This technical solution aims to improve the signal's load-carrying capacity, suppress common-mode interference, and filter out high-frequency noise. However, in a complex vehicle image acquisition and processing environment, noise sources such as electromagnetic interference, signal amplification drift, and power frequency noise affect the purity of the image signal through different pathways. These interferences not only reduce the quality of the image data, but may also have a negative impact on the training and reasoning process of the deep learning model, resulting in a decrease in detection accuracy.
[0005] In view of the above situation, the present invention aims to provide a new solution to solve this problem. Summary of the invention
[0006] In view of the above situation, in order to overcome the defects of the prior art, the purpose of the present invention is to provide an adaptive vehicle vision detection system based on deep learning.
[0007] The technical solution is: an adaptive vehicle visual detection system based on deep learning, including:
[0008] CCD image sensor, used to capture the image of the vehicle to be tested and convert it into analog signal output;
[0009] A signal preprocessing module, used for preprocessing the analog signal output by the CCD image sensor;
[0010] A data processing module, used for converting the pre-processed image signal into a digital signal, and performing high-speed parallel processing on the converted digital signal;
[0011] A host computer, used to receive the digital image data from the data processing module and detect and identify the vehicle using a deep learning model; and
[0012] User interaction interface, used to provide a user operation interface, display test results, and receive user input;
[0013] The signal preprocessing module comprises:
[0014] A preamplifier noise reduction unit, used for preamplifying the output signal of the CCD image sensor quickly and performing high-frequency noise reduction processing;
[0015] A power frequency notch depth adjustment unit, used for performing power frequency notch processing on the output signal of the preamplifier noise reduction unit, and adding depth feedback adjustment during the power frequency notch process to optimize the quality factor of the notch filter; and
[0016] The photoelectric isolation noise removal unit is used to further eliminate the noise in the output signal of the power frequency notch depth conditioning unit to ensure the purity and stability of the image signal output.
[0017] Furthermore, the preamplifier noise reduction unit includes:
[0018] A differential signal amplification circuit, used for differentially amplifying the output signal of the CCD image sensor to suppress the common mode component on the signal transmission line;
[0019] a reference circuit, electrically connected to the differential signal amplification circuit, for providing a bias reference voltage for the differential amplification process; and
[0020] The high-frequency suppression circuit is used to suppress the high-frequency components in the output signal of the differential-mode signal amplification circuit to filter out the high-frequency noise in the image signal.
[0021] Furthermore, the differential signal amplification circuit includes an op amp AR1 and a transistor VT1, the in-phase input terminal of the op amp AR1 is connected to resistor R1, resistor R5 and one end of capacitor C1 through resistor R3, the other end of resistor R1 is connected to the positive signal output interface of the CCD image sensor, the other end of capacitor C1 is grounded, the other end of resistor R5 is connected to the emitter of transistor VT1, the inverting input terminal of the op amp AR1 is connected to resistor R2 and one end of capacitor C2 through resistor R4, the other end of resistor R2 is connected to the negative signal output interface of the CCD image sensor, the other end of capacitor C2 is grounded, and the output terminal of the op amp AR1 is connected to the base of transistor VT1.
[0022] Furthermore, the reference circuit includes a voltage regulator diode DZ1, an op amp AR2 and a transistor VT2, the cathode of the voltage regulator diode DZ1 is connected to a +5V power supply and one end of a resistor R6, the anode of the voltage regulator diode DZ1 is connected to the in-phase input terminal of the op amp AR2 and is grounded through a resistor R8, the other end of the resistor R6 is connected to the emitter of the transistor VT2 and the inverting input terminal of the op amp AR2, the output terminal of the op amp AR2 is connected to the base of the transistor VT2 through a resistor R7, and the collector of the transistor VT2 is connected to one end of a resistor R5.
[0023] Furthermore, the high-frequency suppression circuit includes a MOS tube Q1 and a MOS tube Q2, the drain of the MOS tube Q1 is connected to the cathode of the diode VD1, and the gate of the MOS tube Q1 and the drain of the MOS tube Q2 are connected through a parallel resistor R11 and a capacitor C3, the anode of the diode VD1 is connected to the collector of the transistor VT1 through an inductor L1, the gate of the MOS tube Q2 is connected to the cathode of the voltage-stabilizing diode DZ2, and the output end of the operational amplifier AR1 and one end of the resistor R9 are connected through a parallel resistor R10 and a capacitor C4, the source of the MOS tube Q2, the anode of the voltage-stabilizing diode DZ2 and the other end of the resistor R9 are grounded, the source of the MOS tube Q1 is connected to one end of the resistor R12 and the input end of the power frequency notch depth conditioning unit, and the other end of the resistor R12 is grounded through a capacitor C5.
[0024] Furthermore, the power frequency notch depth adjustment unit includes:
[0025] A double-T-type notch filter, used for performing power frequency notching on the signal processed by the preamplifier noise reduction unit to reduce power frequency noise in the image; and
[0026] The depth feedback adjustment circuit is used to adjust the notch depth and bandwidth of the double-T notch filter.
[0027] Furthermore, the double-T-type trap includes an RC trap network and an op amp AR3, the RC trap network includes resistors R13, R14, R15 and capacitors C6, C7, C8, one end of the resistor R13 and the capacitor C6 are connected to the output end of the preamplifier noise reduction unit, the other end of the resistor R13 is connected to the resistor R14 and one end of the capacitor C8, the other end of the capacitor C6 is connected to the resistor R15 and one end of the capacitor C7, the other end of the resistor R14 and the capacitor C7 is connected to the in-phase input end of the op amp AR3, the other end of the resistor R15 and the capacitor C8 is connected to the output end of the depth feedback regulation circuit, the inverting input end of the op amp AR3 is grounded through the resistor R16, and the output end of the op amp AR3 and the input end of the depth feedback regulation circuit are connected through the resistor R17.
[0028] Furthermore, the depth feedback regulation circuit includes an op amp AR4, a transistor VT3 and a variable resistor RP1, the inverting input terminal of the op amp AR4 is connected to the output terminal of the op amp AR3 through a resistor R18, and is grounded through a resistor R19, the non-inverting input terminal of the op amp AR4 is grounded, the output terminal of the op amp AR4 is connected to the base of the transistor VT3, the emitter of the transistor VT3 is connected to one end of the capacitor C9 and the inverting input terminal of the op amp AR4, the collector of the transistor VT3 is connected to the other end of the capacitor C9 and the regulation end of the variable resistor RP1, the other end of the variable resistor RP1 is connected to the resistor R15 and the other end of the capacitor C8, and is grounded through a resistor R20.
[0029] Further, the photoelectric isolation noise removal unit includes a photoelectric coupler and a spike removal circuit, the anode of the transmitting end of the photoelectric coupler is connected to the output end of the operational amplifier AR3 through a resistor R21, the collector of the receiving end of the photoelectric coupler is connected to a +5V power supply, and the emitter of the receiving end of the photoelectric coupler is connected to the spike removal circuit;
[0030] The spike removal circuit includes a MOS tube Q3, the gate of the MOS tube Q3 is connected to the receiving end emitter of the photoelectric coupler and is grounded through a resistor R22, the drain of the MOS tube Q3 is connected to the capacitor C10, one end of the inductor L2 and the anode of the diode VD2, the other end of the capacitor C10 and the cathode of the diode VD2 are connected to the other end of the inductor L2 and a +5V power supply through a resistor R23, the source of the MOS tube Q3 is connected to one end of the resistor R24, the cathode of the voltage stabilizing diode DZ3 and the data processing module, and the other end of the resistor R24 and the anode of the voltage stabilizing diode DZ3 are grounded.
[0031] Furthermore, the data processing module includes:
[0032] An analog-to-digital converter, used for converting the pre-processed analog image signal into a digital signal;
[0033] The field programmable gate array is used to perform calculation processing on the digital signal and send the processed data to the host computer.
[0034] Through the above technical scheme, the beneficial effects of the present invention are as follows: the adaptive vehicle visual detection system based on deep learning of the present invention realizes multiple optimization processing of image signals through the circuit design of the signal preprocessing module. The signal preprocessing module effectively reduces the influence of various noise sources including electromagnetic interference, high-frequency noise, and power frequency noise on the image quality by integrating the pre-amplification noise reduction unit, the power frequency notch depth conditioning unit, and the photoelectric isolation noise removal unit, and significantly improves the purity and stability of the image signal, which not only optimizes the input quality of the image data, but also enhances the adaptability and detection accuracy of the deep learning model in complex and changeable environments. Therefore, the adaptive vehicle visual detection system of the present invention can achieve higher-precision vehicle visual detection, and provides strong technical support and guarantee for applications in the fields of intelligent transportation, autonomous driving, etc. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] Figure 1 It is a structural block diagram of the system module of the present invention.
[0036] Figure 2 It is a structural block diagram of the signal preprocessing module in the present invention.
[0037] Figure 3 The figure is a circuit diagram of the preamplifier noise reduction unit in the present invention.
[0038] Figure 4 The circuit diagram of the power frequency notch depth conditioning unit in the present invention.
[0039] Figure 5 This is a circuit schematic diagram of the photoelectric isolation and noise removal unit in the present invention. DETAILED DESCRIPTION
[0040] The above and other technical contents, features and effects of the present invention are described in detail below with reference to the attached Figure 1 To Attachment Figure 5 The detailed description of the embodiments will clearly show that the structural contents mentioned in the following embodiments are all based on the drawings in the specification.
[0041] Exemplary embodiments of the present invention will be described below with reference to the accompanying drawings.
[0042] like Figure 1 As shown, the adaptive vehicle visual detection system based on deep learning includes:
[0043] CCD image sensor, used to capture the image of the vehicle to be tested and convert it into analog signal output;
[0044] A signal preprocessing module, used for preprocessing the analog signal output by the CCD image sensor;
[0045] A data processing module, used for converting the pre-processed image signal into a digital signal, and performing high-speed parallel processing on the converted digital signal;
[0046] A host computer, used to receive digital image data from the data processing module and detect and identify vehicles using a deep learning model; and
[0047] The user interaction interface is used to provide a user operation interface, display test results, and receive user input.
[0048] In the above, in order to optimize the processing effect of the image signal and improve the visual detection accuracy, the signal preprocessing module specifically includes:
[0049] The preamplifier noise reduction unit 10 is used to preamplify the output signal of the CCD image sensor quickly and perform high-frequency noise reduction processing;
[0050] The power frequency notch depth adjustment unit 20 is used to perform power frequency notch processing on the output signal of the preamplifier noise reduction unit 10, and add depth feedback adjustment during the power frequency notch process to optimize the quality factor of the notch filter; and
[0051] The photoelectric isolation noise removal unit 30 is used to further eliminate the noise in the output signal of the power frequency notch depth conditioning unit 20 to ensure the purity and stability of the image signal output.
[0052] like Figure 2 and 3 As shown, in the specific implementation process, in order to improve the amplification and filtering effect of the weak signal output by the CCD image sensor, the specific composition of the pre-amplification noise reduction unit 10 includes:
[0053] The differential signal amplifying circuit 101 is used to differentially amplify the output signal of the CCD image sensor to suppress the common mode component on the signal transmission line;
[0054] A reference circuit 102 is electrically connected to the differential signal amplification circuit 101 and is used to provide a bias reference voltage for the differential amplification process; and
[0055] The high frequency suppression circuit 103 is used to suppress the high frequency component in the output signal of the differential signal amplification circuit 101 to filter out the high frequency noise in the image signal.
[0056] Among them, the differential signal amplification circuit includes an operational amplifier AR1 and a transistor VT1, the in-phase input terminal of the operational amplifier AR1 is connected to the resistor R1, the resistor R5 and one end of the capacitor C1 through the resistor R3, the other end of the resistor R1 is connected to the positive signal output interface of the CCD image sensor, the other end of the capacitor C1 is grounded, the other end of the resistor R5 is connected to the emitter of the transistor VT1, the inverting input terminal of the operational amplifier AR1 is connected to the resistor R2 and one end of the capacitor C2 through the resistor R4, the other end of the resistor R2 is connected to the negative signal output interface of the CCD image sensor, the other end of the capacitor C2 is grounded, and the output terminal of the operational amplifier AR1 is connected to the base of the transistor VT1.
[0057] The reference circuit includes a voltage regulator diode DZ1, an operational amplifier AR2 and a transistor VT2. The cathode of the voltage regulator diode DZ1 is connected to a +5V power supply and one end of a resistor R6. The anode of the voltage regulator diode DZ1 is connected to the non-inverting input terminal of the operational amplifier AR2 and is grounded through a resistor R8. The other end of the resistor R6 is connected to the emitter of the transistor VT2 and the inverting input terminal of the operational amplifier AR2. The output terminal of the operational amplifier AR2 is connected to the base of the transistor VT2 through a resistor R7. The collector of the transistor VT2 is connected to one end of a resistor R5.
[0058] The high-frequency suppression circuit includes a MOS tube Q1 and a MOS tube Q2. The drain of the MOS tube Q1 is connected to the cathode of the diode VD1, and the gate of the MOS tube Q1 and the drain of the MOS tube Q2 are connected through a parallel resistor R11 and a capacitor C3. The anode of the diode VD1 is connected to the collector of the transistor VT1 through an inductor L1. The gate of the MOS tube Q2 is connected to the cathode of the voltage-stabilizing diode DZ2, and is connected to the output end of the operational amplifier AR1 and one end of the resistor R9 through a parallel resistor R10 and a capacitor C4. The source of the MOS tube Q2, the anode of the voltage-stabilizing diode DZ2 and the other end of the resistor R9 are grounded. The source of the MOS tube Q1 is connected to one end of the resistor R12 and the input end of the power frequency notch depth conditioning unit, and the other end of the resistor R12 is grounded through a capacitor C5.
[0059] In the above, the specific working principle and process of the preamplifier noise reduction unit are as follows: First, the differential signal amplification circuit uses a dual-channel RC filter to perform low-pass noise reduction on the differential output signal of the CCD image sensor, and then sends it to the operational amplifier AR1 for differential amplification processing to suppress the common-mode component on the signal transmission line; at the same time, the transistor VT1 plays a follow-up amplification role at the output end of the operational amplifier AR1, which quickly increases the image signal amplitude while reducing the influence of common-mode noise. The reference circuit provides a bias reference voltage to the non-inverting input terminal of the operational amplifier AR1 during the differential amplification process, wherein the operational amplifier AR2 and the transistor VT2 form a constant current source circuit, and a stable comparison voltage is formed at the non-inverting input terminal of the operational amplifier AR2 through the voltage stabilizing diode DZ1. The comparator principle is used to make the transistor VT2 output a stable bias reference voltage at its collector, and the bias reference voltage is used to offset the influence of temperature changes or component parameter changes on the working state of the operational amplifier AR1, ensuring that the differential signal amplification circuit can work stably under different environmental conditions; at the same time, the bias reference voltage makes the operational amplifier AR1 work in its linear region, avoiding distortion in the image signal amplification process.
[0060] Secondly, a high-frequency suppression circuit is provided to suppress the high-frequency components in the output signal of the differential-mode signal amplifier circuit. Specifically, the LD-RC filter formed by the inductor L1, the diode VD1, the resistor R11 and the capacitor C3 filters the high-frequency noise in the output signal of the differential-mode signal amplifier circuit, and the MOS transistor Q1 and the MOS transistor Q2 form a composite regulating tube, wherein the MOS transistor Q1 serves as a main amplifier to receive the filtered image signal, and the MOS transistor Q2 serves as a regulating tube to stabilize and regulate the working process of the MOS transistor Q1. Specifically, through the RC buffering effect of the resistor R10 and the capacitor C4, and the amplitude stabilizing effect of the voltage stabilizing diode DZ2, a stable conduction voltage is provided for the gate of the MOS transistor Q2, thereby ensuring the stability of the conduction work of the MOS transistor Q2, so as to eliminate the influence of the self-excited interference generated by the LD-RC filter during the working process on the MOS transistor Q1, and then ensure the stability of the output signal of the MOS transistor Q1.
[0061] In actual applications, power frequency noise can easily cause stripes or ripples in the image during vehicle visual detection, which can introduce misleading features into the deep learning model and make it difficult to correctly learn and identify vehicle features. Therefore, a power frequency notch depth conditioning unit is set at the back end of the preamplifier noise reduction unit to suppress power frequency noise. Specifically, Figure 2 and 4 As shown, the power frequency notch depth adjustment unit 20 includes:
[0062] A double T-type notch filter, used for performing power frequency notching on the signal processed by the preamplifier noise reduction unit to reduce power frequency noise in the image; and
[0063] The depth feedback adjustment circuit is used to adjust the notch depth and bandwidth of the double-T notch filter.
[0064] Among them, the double T-type trap includes an RC trap network and an op amp AR3, the RC trap network includes resistors R13, R14, R15 and capacitors C6, C7, C8, one end of the resistor R13 and the capacitor C6 are connected to the output end of the preamplifier noise reduction unit, the other end of the resistor R13 is connected to the resistor R14 and one end of the capacitor C8, the other end of the capacitor C6 is connected to the resistor R15 and one end of the capacitor C7, the other end of the resistor R14 and the capacitor C7 is connected to the in-phase input end of the op amp AR3, the other end of the resistor R15 and the capacitor C8 is connected to the output end of the depth feedback regulation circuit, the inverting input end of the op amp AR3 is grounded through the resistor R16, and the output end of the op amp AR3 and the input end of the depth feedback regulation circuit are connected through the resistor R17.
[0065] The above-mentioned dual T-type notch filter can effectively suppress power frequency noise in the image signal, making useful image information more prominent, which helps the deep learning model to learn and identify vehicle features more accurately. Due to the complex actual vehicle visual inspection environment, the fixed quality factor (Q value) of the dual T-type notch filter cannot fully adapt to the complex and changing application environment, especially when the noise characteristics vary greatly, which may lead to poor notch effect or excessive notch.
[0066] Therefore, deep feedback regulation is added in the process of power frequency trapping to optimize the Q value of the trap. Specifically, the deep feedback regulation circuit includes an op amp AR4, a transistor VT3 and a variable resistor RP1, the inverting input terminal of the op amp AR4 is connected to the output terminal of the op amp AR3 through a resistor R18, and is grounded through a resistor R19, the non-inverting input terminal of the op amp AR4 is grounded, the output terminal of the op amp AR4 is connected to the base of the transistor VT3, the emitter of the transistor VT3 is connected to one end of the capacitor C9 and the inverting input terminal of the op amp AR4, the collector of the transistor VT3 is connected to the other end of the capacitor C9 and the regulating end of the variable resistor RP1, the other end of the variable resistor RP1 is connected to the resistor R15 and the other end of the capacitor C8, and is grounded through a resistor R20.
[0067] Among them, resistors R18 and R19 use the resistor shunt effect to sample the output signal of op amp AR3, and then send it to the follower amplifier composed of op amp AR4 and transistor VT3 to quickly amplify the sampled signal. The sampling signal amplification process of capacitor C8 plays a compensating and stabilizing role. The amplified sampling signal is fed back to the RC notch network, thereby forming a deep feedback adjustment for the notch filter. The deep feedback adjustment mechanism can dynamically adjust the Q value of the notch filter and optimize the performance of the notch filter. When the noise characteristics change greatly, it helps the notch filter to better adapt to environmental changes and maintain the best noise suppression effect. At the same time, the variable resistor RP1 allows the user to adjust the feedback depth according to actual needs, thereby changing the Q value of the notch filter without redesigning or replacing hardware. This flexibility enables the notch filter to adapt to different noise characteristics and environmental conditions.
[0068] In order to ensure the effective output of the image signal, the image signal after the power frequency notch processing is sent to the photoelectric isolation noise removal unit for further processing. Figure 2 and 5As shown, the photoelectric isolation noise removal unit 30 includes a photoelectric coupler U1 and a spike removal circuit, wherein the anode of the transmitting end of the photoelectric coupler U1 is connected to the output end of the operational amplifier AR3 through a resistor R21, the collector of the receiving end of the photoelectric coupler U1 is connected to a +5V power supply, and the emitter of the receiving end of the photoelectric coupler U1 is connected to the spike removal circuit; wherein the photoelectric coupler U1 transmits electrical signals through light as a medium, thereby realizing electrical isolation between input and output, and this isolation method can cut off electromagnetic interference in the image signal transmission path, thereby improving the signal-to-noise ratio.
[0069] The spike removal circuit further enhances the isolation effect of the signal path, and specifically includes a MOS tube Q3, wherein the gate of the MOS tube Q3 is connected to the emitter of the receiving end of the photocoupler U1 and is grounded through a resistor R22, the drain of the MOS tube Q3 is connected to the capacitor C10, one end of the inductor L2 and the anode of the diode VD2, the other end of the capacitor C10 and the cathode of the diode VD2 are connected to the other end of the inductor L2 and the +5V power supply through a resistor R23, the source of the MOS tube Q3 is connected to one end of the resistor R24, the cathode of the voltage-stabilizing diode DZ3 and the data processing module, and the other end of the resistor R24 and the anode of the voltage-stabilizing diode DZ3 are grounded.
[0070] Among them, the gate of MOS tube Q3 receives the signal from the photocoupler U1. When the signal contains high-frequency spike noise or circuit self-excitation interference, the inductor and capacitor components in the spike removal circuit will hinder the high-frequency noise and effectively suppress the transmission of the spike noise; at the same time, the clamping effect of the diode is used to prevent the components in the circuit from being damaged by overvoltage. Through the processing of the spike removal circuit, the image signal becomes smoother and more stable, reducing the impact of noise on image quality, which is beneficial to subsequent data processing and image reconstruction.
[0071] The data processing module further digitizes and analyzes the pre-processed image signal, including:
[0072] an analog-to-digital converter (ADC), used to convert the pre-processed analog image signal into a digital signal;
[0073] The field programmable gate array is used to perform computational processing on digital signals, such as image enhancement, edge detection, feature extraction, etc. The processed image data contains feature information for vehicle detection and identification, and is sent to the host computer for further analysis.
[0074] As the core processing unit of the system, the host computer receives digital image data from the FPGA and uses a pre-trained deep learning model to perform complex image analysis and processing. The deep learning model can automatically learn and identify vehicles of different types, angles, and lighting conditions by training a large amount of vehicle image data, and finally output the vehicle detection results, including the vehicle's location, type, speed and other information. This information will be used in subsequent application scenarios, such as traffic monitoring, autonomous driving, etc., thereby achieving accurate monitoring and control of vehicle behavior.
[0075] The host computer also passes the processing results to the user interface, which immediately displays these test results to the user. At the same time, the user interface also provides an intuitive operation interface, allowing users to interact with the system by clicking, sliding or inputting, such as adjusting test parameters, viewing history or exporting test results, etc., thus providing users with a convenient and intuitive way to obtain and use vehicle test information.
[0076] In summary, the adaptive vehicle visual detection system based on deep learning of the present invention realizes multiple optimization processing of image signals through the circuit design of the signal preprocessing module. The signal preprocessing module integrates a pre-amplification noise reduction unit, a power frequency notch depth conditioning unit and a photoelectric isolation noise removal unit, which effectively reduces the impact of various noise sources including electromagnetic interference, high-frequency noise, and power frequency noise on the image quality, and significantly improves the purity and stability of the image signal. It not only optimizes the input quality of the image data, but also enhances the adaptability and detection accuracy of the deep learning model in complex and changeable environments. Therefore, the adaptive vehicle visual detection system of the present invention can achieve higher-precision vehicle visual detection, and provides strong technical support and guarantee for applications in the fields of intelligent transportation, autonomous driving, etc.
[0077] The above is a further detailed description of the present invention in combination with a specific implementation method, and it cannot be determined that the specific implementation of the present invention is limited to this; for technical personnel in the technical field to which the present invention belongs and related technical fields, based on the technical solution of the present invention, the expansion and replacement of operating methods and data should all fall within the protection scope of the present invention.
Claims
1. Adaptive vehicle visual detection system based on deep learning, characterized by: include: CCD image sensor, used to capture the image of the vehicle to be tested and convert it into analog signal output; A signal preprocessing module, used for preprocessing the analog signal output by the CCD image sensor; A data processing module, used for converting the pre-processed image signal into a digital signal, and performing high-speed parallel processing on the converted digital signal; A host computer, used to receive the digital image data from the data processing module and detect and identify the vehicle using a deep learning model; as well as User interaction interface, used to provide a user operation interface, display test results, and receive user input; The signal preprocessing module comprises: A preamplifier noise reduction unit, used for preamplifying the output signal of the CCD image sensor quickly and performing high-frequency noise reduction processing; A power frequency notch depth adjustment unit, used for performing power frequency notch processing on the output signal of the preamplifier noise reduction unit, and adding depth feedback adjustment during the power frequency notch process to optimize the quality factor of the notch filter; and The photoelectric isolation noise removal unit is used to further eliminate the noise in the output signal of the power frequency notch depth conditioning unit to ensure the purity and stability of the image signal output.
2. The deep learning-based adaptive vehicle visual detection system according to claim 1, characterized in that: The preamplifier noise reduction unit comprises: A differential signal amplification circuit, used for differentially amplifying the output signal of the CCD image sensor to suppress the common mode component on the signal transmission line; a reference circuit, electrically connected to the differential signal amplification circuit, for providing a bias reference voltage for the differential amplification process; and The high-frequency suppression circuit is used to suppress the high-frequency components in the output signal of the differential-mode signal amplification circuit to filter out the high-frequency noise in the image signal.
3. The deep learning-based adaptive vehicle visual detection system according to claim 2, characterized in that: The differential signal amplification circuit includes an operational amplifier AR1 and a transistor VT1. The in-phase input terminal of the operational amplifier AR1 is connected to the resistor R1, the resistor R5 and one end of the capacitor C1 through the resistor R3. The other end of the resistor R1 is connected to the positive signal output interface of the CCD image sensor. The other end of the capacitor C1 is grounded. The other end of the resistor R5 is connected to the emitter of the transistor VT1. The inverting input terminal of the operational amplifier AR1 is connected to the resistor R2 and one end of the capacitor C2 through the resistor R4. The other end of the resistor R2 is connected to the negative signal output interface of the CCD image sensor. The other end of the capacitor C2 is grounded. The output terminal of the operational amplifier AR1 is connected to the base of the transistor VT1.
4. The deep learning-based adaptive vehicle visual detection system according to claim 3, characterized in that: The reference circuit includes a voltage stabilizing diode DZ1, an operational amplifier AR2 and a transistor VT2, wherein the cathode of the voltage stabilizing diode DZ1 is connected to a +5V power supply and one end of a resistor R6, the anode of the voltage stabilizing diode DZ1 is connected to the in-phase input end of the operational amplifier AR2 and is grounded through a resistor R8, the other end of the resistor R6 is connected to the emitter of the transistor VT2 and the inverting input end of the operational amplifier AR2, the output end of the operational amplifier AR2 is connected to the base of the transistor VT2 through a resistor R7, and the collector of the transistor VT2 is connected to one end of a resistor R5.
5. The deep learning-based adaptive vehicle visual detection system according to claim 4, characterized in that: The high-frequency suppression circuit includes a MOS tube Q1 and a MOS tube Q2, the drain of the MOS tube Q1 is connected to the cathode of the diode VD1, and the gate of the MOS tube Q1 and the drain of the MOS tube Q2 are connected through a parallel resistor R11 and a capacitor C3, the anode of the diode VD1 is connected to the collector of the transistor VT1 through an inductor L1, the gate of the MOS tube Q2 is connected to the cathode of the voltage-stabilizing diode DZ2, and the output end of the operational amplifier AR1 and one end of the resistor R9 are connected through a parallel resistor R10 and a capacitor C4, the source of the MOS tube Q2, the anode of the voltage-stabilizing diode DZ2 and the other end of the resistor R9 are grounded, the source of the MOS tube Q1 is connected to one end of the resistor R12 and the input end of the power frequency notch depth conditioning unit, and the other end of the resistor R12 is grounded through a capacitor C5.
6. The deep learning-based adaptive vehicle visual detection system according to claim 5, characterized in that: The power frequency notch depth adjustment unit comprises: A double-T-type notch filter, used for performing power frequency notching on the signal processed by the preamplifier noise reduction unit to reduce power frequency noise in the image; and The depth feedback adjustment circuit is used to adjust the notch depth and bandwidth of the double-T notch filter.
7. The deep learning-based adaptive vehicle visual detection system according to claim 6, characterized in that: The double-T type notch filter includes an RC notch network and an op amp AR3, wherein the RC notch network includes resistors R13, R14, R15 and capacitors C6, C7, and C8, wherein one end of the resistor R13 and the capacitor C6 are connected to the output end of the preamplifier noise reduction unit, the other end of the resistor R13 is connected to one end of the resistor R14 and the capacitor C8, the other end of the capacitor C6 is connected to one end of the resistor R15 and the capacitor C7, the other end of the resistor R14 and the capacitor C7 is connected to the in-phase input end of the op amp AR3, the other end of the resistor R15 and the capacitor C8 is connected to the output end of the depth feedback regulation circuit, the inverting input end of the op amp AR3 is grounded through a resistor R16, and the output end of the op amp AR3 and the input end of the depth feedback regulation circuit are connected through a resistor R17.
8. The deep learning-based adaptive vehicle visual detection system according to claim 7, characterized in that: The depth feedback regulation circuit includes an op amp AR4, a transistor VT3 and a variable resistor RP1, the inverting input terminal of the op amp AR4 is connected to the output terminal of the op amp AR3 through a resistor R18, and is grounded through a resistor R19, the non-inverting input terminal of the op amp AR4 is grounded, the output terminal of the op amp AR4 is connected to the base of the transistor VT3, the emitter of the transistor VT3 is connected to one end of the capacitor C9 and the inverting input terminal of the op amp AR4, the collector of the transistor VT3 is connected to the other end of the capacitor C9 and the regulation end of the variable resistor RP1, the other end of the variable resistor RP1 is connected to the resistor R15 and the other end of the capacitor C8, and is grounded through a resistor R20.
9. The deep learning-based adaptive vehicle visual detection system according to claim 8, characterized in that: The photoelectric isolation and noise removal unit includes a photoelectric coupler and a spike removal circuit, the anode of the transmitting end of the photoelectric coupler is connected to the output end of the operational amplifier AR3 through a resistor R21, the collector of the receiving end of the photoelectric coupler is connected to a +5V power supply, and the emitter of the receiving end of the photoelectric coupler is connected to the spike removal circuit; The spike removal circuit includes a MOS tube Q3, the gate of the MOS tube Q3 is connected to the receiving end emitter of the photoelectric coupler and is grounded through a resistor R22, the drain of the MOS tube Q3 is connected to the capacitor C10, one end of the inductor L2 and the anode of the diode VD2, the other end of the capacitor C10 and the cathode of the diode VD2 are connected to the other end of the inductor L2 and a +5V power supply through a resistor R23, the source of the MOS tube Q3 is connected to one end of the resistor R24, the cathode of the voltage stabilizing diode DZ3 and the data processing module, and the other end of the resistor R24 and the anode of the voltage stabilizing diode DZ3 are grounded.
10. The deep learning-based adaptive vehicle visual detection system according to claim 1 or 9, characterized in that: The data processing module comprises: An analog-to-digital converter, used for converting the pre-processed analog image signal into a digital signal; The field programmable gate array is used to perform calculation processing on the digital signal and send the processed data to the host computer.
Citation Information
Patent Citations
Signal processing circuit of linear array CCD detector
CN108681270A
CCD data high-speed acquisition device based on FPGA
CN220554076U
Cited By
Steel cylinder three-code information online identification system and equipment
CN119167959A
