NIR-Ⅱ fluorescence early detection device and method for citrus huanglongbing based on FPGA edge AI

Abstract

The application discloses a citrus Huanglongbing NIR-II fluorescence early detection device and method based on FPGA edge AI, and belongs to the technical field of intelligent plant disease detection. The device comprises a NIR-II activatable H2O2 fluorescence nanosensor, an optical acquisition module, a FPGA core processing unit, a man-machine interaction module and a power supply module. The FPGA is internally integrated with an image preprocessing hardware IP core and a lightweight AI inference hardware accelerator, so that pure hardware offline real-time inference is realized. The method takes H2O2 as a specific early target point, activates a response through a NIR-II fluorescence signal, and completes high-speed and low-power consumption detection by the FPGA. The application can give an early warning 15-20 days before the Huanglongbing disease is manifested, and the detection of a single plant is less than or equal to 300 ms, the power consumption is less than 1 W, offline operation is realized, high anti-interference is achieved, and the application is suitable for large-scale nondestructive early detection of citrus in the field, thereby solving the problem that the prior art cannot realize early, real-time, specific and portable detection.

Description

Technical Field

[0001] This invention belongs to the field of intelligent detection technology for plant diseases, specifically relating to an early detection device and method for NIR-II fluorescence of citrus Huanglongbing based on FPGA edge AI. Background Technology

[0002] Citrus Huanglongbing (HLB), caused by the phloem bacterium CLas, is the most devastating disease in the citrus industry. Existing detection methods have significant limitations: 1) qPCR requires laboratory DNA extraction, which is complex, costly, and incapable of rapid field detection; 2) Visual identification relies on leaf mottling and yellowing symptoms, indicating the disease has progressed to a mid-to-late stage; 3) Traditional optical detection lacks specific targets, is easily interfered with by chlorophyll autofluorescence, and cannot output results in real time; 4) General-purpose edge computing platforms have high power consumption, poor real-time performance, and weak anti-interference capabilities, making them unsuitable for prolonged handheld operation in the field. Research has confirmed that CLas infection in citrus induces a significant accumulation of H2O2 in the phloem, with concentration positively correlated with the degree of infection. This concentration abnormally increases 15–20 days before visible symptoms appear, making it an ideal specific molecular marker for early detection of HLB. Near-infrared II (NIR-II, 1000–1700 nm) fluorescence can effectively avoid interference from plant autofluorescence, achieving high signal-to-noise ratio detection in vivo. FPGAs possess advantages such as parallel computing, low latency, low power consumption, hardware integration, and strong anti-interference capabilities, making them an excellent hardware platform for realizing real-time offline AI inference in the field. Currently, existing technologies cannot achieve early, specific, non-destructive, real-time, low-power, offline detection of citrus Huanglongbing (HLB). Summary of the Invention

[0003] To address the limitations of existing technologies in achieving early, specific, non-destructive, real-time, low-power, and offline detection of citrus Huanglongbing (HLB), this invention proposes a NIR-II fluorescence early detection device and method for HLB based on FPGA edge AI. This device and method enable early warning during the asymptomatic period, second-level detection of individual plants, offline operation, high anti-interference capabilities, and low power consumption for large-scale field screening.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: An early detection device for NIR-II fluorescence of citrus Huanglongbing (HLB) based on FPGA edge AI includes a NIR-II activatable H2O2 fluorescent nanosensor, an optical acquisition module, an FPGA core processing unit, a human-computer interaction module, and a power supply module. The NIR-II activatable H2O2 fluorescent nanosensor is electrically connected to the optical acquisition module. The FPGA core processing unit is electrically connected to the optical acquisition module, the human-computer interaction module, and the power supply module. The FPGA core processing unit integrates an image preprocessing hardware IP core and a lightweight AI inference hardware accelerator to directly complete the acquisition, preprocessing, feature extraction, and disease classification inference of NIR-II fluorescence signals.

[0005] Preferably, the NIR-II activatable H2O2 fluorescent nanosensor is an AIE (Activated Electron Ion) nanosensor. 1035 NPs@Mo / Cu-POM are composed of NIR-II aggregation-induced emission (AIE) fluorophores and H2O2-responsive Mo / Cu-POM quenchers through electrostatic self-assembly. The NIR-II-activated H2O2 fluorescent nanosensor exhibits a weakened quenching effect upon encountering H2O2, and its fluorescence is specifically activated and illuminated.

[0006] Preferably, the optical acquisition module includes an 808nm excitation light source, a long-pass filter of 900nm or above, and an InGaAsNIR photoelectric sensor.

[0007] Preferably, the FPGA core processing unit adopts a pure hardware architecture without a CPU or operating system.

[0008] Preferably, the lightweight AI inference hardware accelerator is a hardware implementation of a lightweight CNN, consisting of a parallel lookup table, adders, and comparator arrays.

[0009] Preferably, the inference delay of a single frame signal is ≤10ms, and the detection time of a single complete plant is ≤300ms.

[0010] Preferably, the total power consumption of the device is less than 1W, and it supports offline operation without network access.

[0011] A method for early detection of NIR-II fluorescence in citrus Huanglongbing (HLB) based on FPGA edge AI, implemented using the aforementioned FPGA edge AI-based early detection device for HLB NIR-II fluorescence, includes the following steps: S1. The NIR-II activatable H2O2 fluorescent nanosensor is introduced into the phloem of citrus petioles or young shoots through needleless permeation. S2. NIR-II fluorescence signals are acquired through an optical acquisition module using 808nm laser excitation. S3. Input the fluorescence signal into the FPGA core processing unit, and complete the noise reduction, normalization and ROI extraction through the image preprocessing hardware IP core; S4. Complete the inference and classification of citrus Huanglongbing through the lightweight AI inference hardware accelerator inside the FPGA, and transmit the inference and classification results to the human-computer interaction module. S5, the human-computer interaction module outputs three levels of results on-site: healthy, early infection, and symptomatic infection.

[0012] Preferably, early detection can be achieved 15–20 days before the visible symptoms of citrus Huanglongbing appear.

[0013] Application of the device or method in field screening, origin clearance testing, and early warning of citrus Huanglongbing (HLB).

[0014] After adopting the above technical solution, the present invention has the following beneficial effects: 1) Early warning: Detection can be achieved 15-20 days before the appearance of visible symptoms of Huanglongbing; 2) High specificity: Using H2O2 as a specific target, it is not affected by drought, nutrient deficiency, or other diseases; 3) FPGA hardware acceleration: Inference latency ≤10ms, single plant detection ≤300ms; 4) Low power consumption: The power consumption of the whole machine is <1W, which is suitable for long-term operation of handheld portable devices; 5) Offline real-time: No network or cloud required, results are output in seconds on site; 6) High anti-interference: Avoids chlorophyll autofluorescence and adapts to strong light, high temperature, and dusty field environments; 7) Non-destructive testing: In-situ detection of living organisms, without damage, and without affecting plant growth. Attached Figure Description

[0015] Figure 1 This is a block diagram of the overall structure of the device of the present invention; Figure 2 This is a flowchart of the method of the present invention; Figure 3 The response curves of the NIR-II activated H2O2 fluorescent nanosensor prepared in Example 1 of this invention to different concentrations of H2O2 are shown. Figure 4 This is a comparison of the NIR-II fluorescence intensity of the main vein of healthy, unsymptomatic, and symptomatic citrus leaves in Example 2 of the present invention; Figure 5 This is a comparison of NIR-II fluorescence intensity of citrus leaves under different stress conditions in Example 4 of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0017] like Figures 1 to 5 As shown.

[0018] An early detection device for NIR-II fluorescence of citrus Huanglongbing (HLB) based on FPGA edge AI includes a NIR-II activatable H2O2 fluorescent nanosensor, an optical acquisition module, an FPGA core processing unit, a human-computer interaction module, and a power supply module. The NIR-II activatable H2O2 fluorescent nanosensor is electrically connected to the optical acquisition module. The FPGA core processing unit is electrically connected to the optical acquisition module, the human-computer interaction module, and the power supply module. The FPGA core processing unit integrates an image preprocessing hardware IP core and a lightweight AI inference hardware accelerator to directly complete the acquisition, preprocessing, feature extraction, and disease classification inference of NIR-II fluorescence signals.

[0019] The NIR-II-activated H2O2 fluorescent nanosensor is an AIE (Activated Electron Microscopy) 1035 NPs@Mo / Cu-POM are composed of NIR-II aggregation-induced emission (AIE) fluorophores and H2O2-responsive Mo / Cu-POM quenchers through electrostatic self-assembly. The NIR-II-activated H2O2 fluorescent nanosensor exhibits a weakened quenching effect upon encountering H2O2, and its fluorescence is specifically activated and illuminated.

[0020] The optical acquisition module includes an 808nm excitation light source, a long-pass filter of 900nm or above, and an InGaAsNIR photoelectric sensor.

[0021] The FPGA core processing unit adopts a pure hardware architecture without a CPU or operating system.

[0022] The lightweight AI inference hardware accelerator is a hardware implementation of lightweight CNN, consisting of a parallel lookup table, adders, and comparator arrays.

[0023] The inference delay for a single frame signal is ≤10ms, and the detection time for a single complete plant is ≤300ms.

[0024] The total power consumption of the device is less than 1W, and it supports offline operation without network access.

[0025] A method for early detection of NIR-II fluorescence in citrus Huanglongbing (HLB) based on FPGA edge AI, implemented using the aforementioned FPGA edge AI-based early detection device for HLB NIR-II fluorescence, includes the following steps: S1. The NIR-II activatable H2O2 fluorescent nanosensor is introduced into the phloem of citrus petioles or young shoots through needleless permeation. S2. NIR-II fluorescence signals are acquired through an optical acquisition module using 808nm laser excitation. S3. Input the fluorescence signal into the FPGA core processing unit, and complete the noise reduction, normalization and ROI extraction through the image preprocessing hardware IP core; S4. Complete the inference and classification of citrus Huanglongbing through the lightweight AI inference hardware accelerator inside the FPGA, and transmit the inference and classification results to the human-computer interaction module. S5, the human-computer interaction module outputs three levels of results on-site: healthy, early infection, and symptomatic infection.

[0026] It enables early detection of citrus Huanglongbing 15–20 days before visible symptoms appear.

[0027] Application of the device or method in field screening, origin clearance testing, and early warning of citrus Huanglongbing (HLB). Example 1: Preparation of NIR-II-activated H2O2 fluorescent nanosensors 1) AIE 1035 AIE was obtained by encapsulating dyes in polystyrene nanospheres using a swelling method. 1035 NPs; 2) Synthesis of Mo / Cu-POM polyoxometalates by hydrothermal method; 3) AIE via electrostatic self-assembly 1035 NPs are compounded with Mo / Cu-POM at a mass ratio of 1:10; 4) AIE is prepared. 1035 NPs@Mo / Cu-POM, with a particle size range of approximately 230-240 nm, have a detection limit of 0.22 μM for H2O2 and a response time ≤ 1 min. Example 2: Detection of H2O2 in the main vein of citrus leaves using a NIR-II activated H2O2 fluorescent nanosensor. The NIR-II activatable H2O2 fluorescent nanosensor prepared in Example 1 was prepared into a 0.1 mg / mL suspension using deionized water and introduced into the phloem of new shoots using a needle-free permeation injection device. After standing for 20 minutes, the NIR-II activatable H2O2 fluorescent nanosensor was fully distributed and background fluorescence was quenched. Fluorescence signals were collected using an InGaAsNIR photoelectric sensor. The fluorescence value of the main vein of healthy leaves was 1.05, that of unsymptomatic leaves was 1.82, and that of symptomatic leaves was 2.81, showing a significant difference. Example 3: Deploying a lightweight CNN model on an FPGA The 1×10000 NIR-II fluorescence intensity sequences collected by the InGaAsNIR photoelectric sensor are reassembled into a 100×100 two-dimensional matrix inside the FPGA to form a spatial fluorescence distribution pseudo image. The image preprocessing hardware IP core completes median filtering for noise reduction, maximum and minimum value normalization, and background ROI cropping. A five-layer pure fixed-point lightweight CNN network was designed: the input layer is a 100×100×1 fluorescence intensity matrix; the first convolutional layer uses eight 3×3 depthwise separable convolutional kernels with a stride of 1, padding with SAME, and outputs a 100×100×8 feature map after ReLU activation; then it is downsampled to 50×50×8 by 2×2 max pooling; the second convolutional layer uses sixteen 1×1 pointwise convolutional kernels to output a 50×50×16 feature map; then it is obtained by global average pooling to obtain a 1×1×16 vector; finally, a fully connected layer maps the 16-dimensional features to a 3-dimensional classification output (healthy, early infection, symptomatic infection), and a lookup table method is used to approximate softmax; the total number of parameters in the entire network is approximately 1840 weights, all using 8-bit fixed-point numbers. A Xilinx Spartan-7 XC7S15 FPGA was selected, and a parallel lookup table inference accelerator was built using Verilog to realize pure hardware AI inference without a CPU or operating system. Example 4: Field detection of citrus Huanglongbing (HLB) Asymptomatic new shoots of 3-year-old navel oranges were selected from the field. A 0.1 mg / mL sensor suspension was injected without needles and allowed to stand for 20 minutes. The device was then turned on to excite the main leaf vein with an 808 nm laser and collect NIR-II fluorescence signals. After FPGA preprocessing and inference, the results of healthy plants, Huanglongbing infection, magnesium deficiency, iron deficiency, and zinc deficiency were output, which were consistent with the qPCR verification results.

[0028] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A FPGA-based edge AI-based early detection device for NIR-II fluorescence of citrus Huanglongbing (HLB), characterized in that: The system includes a NIR-II activatable H2O2 fluorescent nanosensor, an optical acquisition module, an FPGA core processing unit, a human-computer interaction module, and a power supply module. The NIR-II activatable H2O2 fluorescent nanosensor is electrically connected to the optical acquisition module. The FPGA core processing unit is electrically connected to the optical acquisition module, the human-computer interaction module, and the power supply module. The FPGA core processing unit integrates an image preprocessing hardware IP core and a lightweight AI inference hardware accelerator, which are used to directly complete the acquisition, preprocessing, feature extraction, and disease classification inference of the NIR-II fluorescence signal.

2. The FPGA-based edge AI-based early detection device for NIR-II fluorescence of citrus Huanglongbing as described in claim 1, characterized in that: The NIR-II-activated H2O2 fluorescent nanosensor is an AIE (Activated Electron Microscopy) 1035 NPs@Mo / Cu-POM are composed of NIR-II aggregation-induced emission (AIE) fluorophores and H2O2-responsive Mo / Cu-POM quenchers through electrostatic self-assembly. The NIR-II-activated H2O2 fluorescent nanosensor exhibits a weakened quenching effect upon encountering H2O2, and its fluorescence is specifically activated and illuminated.

3. The FPGA-based edge AI-based early detection device for NIR-II fluorescence of citrus Huanglongbing as described in claim 1, characterized in that: The optical acquisition module includes an 808nm excitation light source, a long-pass filter of 900nm or above, and an InGaAsNIR photoelectric sensor.

4. The FPGA-based edge AI-based early detection device for NIR-II fluorescence of citrus Huanglongbing as described in claim 1, characterized in that: The FPGA core processing unit adopts a pure hardware architecture without a CPU or operating system.

5. The FPGA-based edge AI-based early detection device for NIR-II fluorescence of citrus Huanglongbing as described in claim 1, characterized in that: The lightweight AI inference hardware accelerator is a hardware implementation of lightweight CNN, consisting of a parallel lookup table, adders, and comparator arrays.

6. The FPGA-based edge AI-based early detection device for NIR-II fluorescence of citrus Huanglongbing as described in claim 1, characterized in that: The inference delay for a single frame signal is ≤10ms, and the detection time for a single complete plant is ≤300ms.

7. The FPGA-based edge AI-based early detection device for NIR-II fluorescence of citrus Huanglongbing as described in claim 1, characterized in that: The total power consumption of the device is less than 1W, and it supports offline operation without network access.

8. A method for early detection of NIR-II fluorescence in citrus Huanglongbing based on FPGA edge AI, characterized in that, The method employs the FPGA-based edge AI-based NIR-II fluorescence early detection device for citrus Huanglongbing (HLB) as described in any one of claims 1-7, comprising the following steps: S1. The NIR-II activatable H2O2 fluorescent nanosensor is introduced into the phloem of citrus petioles or young shoots through needleless permeation. S2. NIR-II fluorescence signals are acquired through an optical acquisition module using 808nm laser excitation. S3. Input the fluorescence signal into the FPGA core processing unit, and complete the noise reduction, normalization and ROI extraction through the image preprocessing hardware IP core; S4. Complete the inference and classification of citrus Huanglongbing through the lightweight AI inference hardware accelerator inside the FPGA, and transmit the inference and classification results to the human-computer interaction module. S5, the human-computer interaction module outputs three levels of results on-site: healthy, early infection, and symptomatic infection.

9. The method for early detection of NIR-II fluorescence in citrus Huanglongbing based on FPGA edge AI as described in claim 8, characterized in that: It enables early detection of citrus Huanglongbing 15–20 days before visible symptoms appear.

10. The application of the device as described in claim 1 or the method as described in claim 8 in field screening, origin clearance testing, and early warning of citrus Huanglongbing (HLB).

Images