Method and system for detecting types of fusarium moniliforme in rice seeds

Abstract

The invention belongs to the technical field of rice seed quality detection, and particularly relates to a method and a system for detecting types of fusarium moniliforme in rice seeds, the method comprises the following steps: mixing the rice seeds to be detected with a composite lysate, purifying DNA through a paramagnetic particle method, introducing the purified DNA into a micro-fluidic chip preloaded with a freeze-drying reagent, performing multiple LAMP amplification by using a primer group, and detecting the types of the fusarium moniliforme in the micro-fluidic chip. The reaction condition is that the constant temperature is 62-65 DEG C for 25-35 minutes, the pathogenic bacteria genus and pathotype are judged through real-time fluorescence signals, a hyperspectral imager is synchronously adopted to collect seed surface images, and the problem of leak detection of a single detection technology in the detection of the rice fusarium moniliforme is effectively solved through the combination of a hyperspectral technology and a convolutional neural network (CNN). The hyperspectral technology can capture rich spectral information of the rice seeds, the convolutional neural network has strong feature learning and mode recognition capabilities, the spectral information can be deeply mined and analyzed, the advantages of the hyperspectral technology and the convolutional neural network are complementary, and the accuracy and reliability of detection are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of rice seed quality detection, and specifically relates to a method and system for detecting the type of bakanae disease bacteria in rice seeds. Background Art

[0002] Bakanae disease of rice is an important rice disease caused by fungi and occurs in all rice production areas in the world. After the bakanae disease bacteria infect rice seeds, it will cause the rice seedlings to grow excessively, be weak and slender, and prone to lodging, seriously affecting the emergence rate and seedling quality of rice. In the adult stage, diseased plants will also show symptoms such as poor growth and reduced seed setting rate, causing serious losses to rice yield and quality.

[0003] Traditional methods for detecting bakanae disease bacteria mainly rely on the isolation and culture of pathogenic bacteria and morphological observation. Although this method is classical, it has many disadvantages. For example, the detection period is long, generally taking several days to several weeks to obtain results, which is difficult to meet the needs of rapid detection; moreover, it has high technical requirements for operators, requiring professional microbiological knowledge and rich experience, and the judgment results of different operators may vary; in addition, morphological identification can only preliminarily determine the type of bacteria, and it is difficult to accurately distinguish some bacteria with similar morphologies and cannot be accurate to the specific type of bakanae disease bacteria.

[0004] Therefore, the present invention provides a method and system for detecting the type of bakanae disease bacteria in rice seeds. Summary of the Invention

[0005] In order to make up for the deficiencies of the prior art and solve at least one of the technical problems proposed in the background art.

[0006] The technical solution adopted by the present invention to solve its technical problems is as follows: A method and system for detecting the type of bakanae disease bacteria in rice seeds according to the present invention includes the following steps:

[0007] A1. Mix the rice seeds to be tested with a composite lysis solution, where the lysis solution contains 0.5% - 1.5% CTAB, 0.1 - 0.3 mg / mL proteinase K, 1% - 3% polyvinylpyrrolidone (PVP), and 0.05% - 0.2% mercaptoethanol, and oscillate and lyse at 60 - 70 °C for 10 - 20 minutes, and purify the DNA by the magnetic bead method;

[0008] A2. Introduce the purified DNA into a microfluidic chip pre-loaded with freeze-dried reagents, perform multiplex LAMP amplification using the primer sets shown in SEQ ID NO: 1 - 6, and the reaction conditions are constant temperature at 62 - 65 °C for 25 - 35 minutes, and determine the genus and pathogenicity of the pathogenic bacteria through real-time fluorescence signals;

[0009] A3. Use a hyperspectral imager to collect seed surface images and input them into a pre-trained two-branch CNN model. The first branch extracts the spatial texture feature convolution layer, and the second branch extracts the spectral sequence feature convolution layer, and outputs the pathogen infection probability and bacterial species classification results.

[0010] A4. Integrate the results of step A1 and step A3 based on the dynamic weight fusion algorithm to generate a comprehensive report including the pathogen type, pathogenicity, infection index (0% to 100%) and confidence level (≥90%).

[0011] The detection threshold of the fluorescence signal amplified by the LAMP is three times the standard deviation of the baseline fluorescence value, and the positive judgment requires the satisfaction of two target genes. The training data of the dual-branch CNN model includes hyperspectral images of healthy seeds, early infected seeds (asymptomatic) and late infected seeds, and random illumination transformation and noise injection data enhancement technology are used during training.

[0012] In the dynamic weight fusion algorithm, if the confidence difference between the molecular detection and spectral analysis results is ≥20%, a manual review prompt is triggered. The hyperspectral imaging module supports batch scanning of multiple seeds and automatically locates the single seed area through an image segmentation algorithm.

[0013] In the dynamic weighted Bayesian network, the initial weight of the molecular detection result is 55% to 65%, the initial weight of the spectral analysis result is 35% to 45%, and the weight ratio is dynamically adjusted with the model confidence.

[0014] The present invention also provides a method for detecting the type of Bakanae bakanae pathogen in rice seeds. The method adopts the above-mentioned system for detecting the type of Bakanae bakanae pathogen in rice seeds, and comprises the following modules:

[0015] S1, nucleic acid extraction module: integrated lysis solution storage tank, microcentrifuge (speed 12000±500rpm) and magnetic bead purification component (magnetic bead particle size 0.1-1μm), supporting single sample processing time ≤15 minutes;

[0016] S2, constant temperature amplification module: using a microfluidic chip (size ≤5cm×3cm×0.5cm) and a semiconductor temperature control unit (accuracy ±0.3℃), the chip contains three independent reaction chambers, which are pre-loaded with freeze-dried primers targeting the RPB2 gene (genus specificity), FfPS1 gene (gibberellin type) and FUM1 gene (toxin type);

[0017] S3, Hyperspectral imaging module: includes near-infrared hyperspectral camera (wavelength range 900-1700nm, spatial resolution 0.2mm / pixel), LED ring light source (color temperature 5000K±200K) and autofocus mechanism, supporting single seed imaging time ≤3 seconds;

[0018] S4, Data Analysis Module:

[0019] (1) Embedded AI processor, running a dual-branch CNN model (model size ≤ 500MB);

[0020] (2) Dynamic weight fusion algorithm;

[0021] (3) User interface (touch screen, supporting Chinese / English / Japanese / Indian languages), capable of generating ISO17025 standard certification reports;

[0022] S5, Communication Module: Supports 4G / 5G / Wi-Fi wireless connections, and synchronizes data to the cloud pathogen database in real time (including spectral and genetic characteristics of ≥ 1000 Fusarium strains).

[0023] The surface of the microfluidic chip is coated with a trehalose-bovine serum albumin (BSA) composite protective layer, which can be stored for more than 6 months at 25°C. The embedded control unit supports the switching between offline and online modes. In the online mode, it can connect to the cloud server through 5G / Wi-Fi to update the pathogen characteristic database in real time.

[0024] The dynamic weight fusion algorithm integrates the results of molecular detection and spectral analysis based on weighted summation. Let the confidence of molecular detection be C1, the confidence of spectral analysis be C2, the weight of molecular detection confidence be W1, with a value range between 55% and 65%, and the weight of spectral analysis be W2, with a value range between 35% and 45%, and W1 + W2 = 1. The calculation formula for the detection confidence is:

[0025] C = C1×W1 + C2×W2

[0026] The image segmentation algorithm starts from a certain seed pixel or region in the image and gradually grows or merges into a complete seed region according to certain similarity criteria (such as spectral feature similarity, color similarity, etc.).

[0027] The embedded control unit supports users to upload private data to the cloud pathogen database and updates the global model parameters through federated learning technology.

[0028] The beneficial effects of the present invention are as follows:

[0029] 1. By combining hyperspectral technology with convolutional neural network (CNN), the problem of missed detection existing in the detection of rice bakanae disease by a single detection technology is effectively overcome. Hyperspectral technology can capture rich spectral information of rice seeds, and the powerful feature learning and pattern recognition capabilities of convolutional neural network can deeply mine and analyze these spectral information. The advantages of the two complement each other, significantly improving the accuracy and reliability of detection.

[0030] 2. By improving the lysis buffer formulation, it can effectively inhibit the interference of impurities such as polysaccharides and phenols in the seeds on the amplification reaction while lysing the seed cells, thus skipping the traditional DNA purification step. This improvement not only simplifies the detection process but also saves detection time, and improves the detection efficiency by saving time compared with the traditional method.

[0031] 3. Through the intelligent adaptive learning ability, it can automatically adjust the detection threshold according to the biological characteristics and spectral feature differences of different rice varieties. When dealing with a rich variety of rice varieties, it can identify the unique features of each variety by learning and analyzing a large amount of sample data, and dynamically optimize the detection parameters to ensure high sensitivity and specificity during the detection of different rice varieties. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The present invention will be further described below with reference to the accompanying drawings.

[0033] Figure 1 is the flowchart of the detection method of a method and system for detecting the type of bakanae disease bacteria in rice seeds according to the present invention;

[0034] Figure 2 is the architecture diagram of the dual-branch CNN model of a method and system for detecting the type of bakanae disease bacteria in rice seeds according to the present invention;

[0035] Figure 3 is the system process planning architecture diagram of a method and system for detecting the type of bakanae disease bacteria in rice seeds according to the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0036] In order to make the technical means, creative features, achieved purposes and effects of the present invention easy to understand, the present invention will be further described below in conjunction with specific embodiments.

[0037] As Figures 1 to 2 shown, the embodiments of the present invention include the following steps:

[0038] A1. Mix the rice seeds to be tested with a composite lysis buffer, the lysis buffer contains 0.5% - 1.5% CTAB, 0.1 - 0.3 mg / mL proteinase K, 1% - 3% polyvinylpyrrolidone (PVP) and 0.05% - 0.2% mercaptoethanol, and oscillate and lyse at 60 - 70 °C for 10 - 20 minutes, and purify the DNA by the magnetic bead method;

[0039] A2. Import the purified DNA into a microfluidic chip pre-loaded with freeze-dried reagents, and perform multiplex LAMP amplification using the primer sets shown in SEQ ID NO: 1 - 6. The reaction conditions are constant temperature at 62 - 65 °C for 25 - 35 minutes, and determine the pathogen genus and pathogenicity through real-time fluorescence signals;

[0040] A3. Simultaneously, a hyperspectral imager is used to collect the surface images of the seeds, which are then input into a pre-trained dual-branch CNN model. The first branch extracts the convolutional layer of spatial texture features, and the second branch extracts the convolutional layer of spectral sequence features, and the output is the probability of pathogen infection and the classification result of the pathogen species.

[0041] A4. Based on the dynamic weight fusion algorithm, the results of step A1 and step A3 are integrated to generate a comprehensive report containing the pathogen type, pathogenicity, infection index (0% - 100%), and confidence level (≥90%).

[0042] The rice seeds to be tested are fully mixed with a composite lysis solution, which consists of multiple key components. The concentration of CTAB (cetyltrimethylammonium bromide) is controlled at 0.5% - 1.5%, which can effectively destroy the cell membrane structure of the seed cells and promote the release of cell contents. The concentration of proteinase K is 0.1 - 0.3 mg / mL, which can degrade proteins and prevent interference with subsequent DNA extraction. 1% - 3% of polyvinylpyrrolidone (PVP) can bind to impurities such as polysaccharides and polyphenols in the seeds to improve the purity of DNA. 0.05% - 0.2% of mercaptoethanol has an antioxidant effect to protect DNA from oxidative degradation. After mixing, the sample is placed in an environment of 60 - 70 °C and lysed at an appropriate oscillation rate for 10 - 20 minutes. After lysis, the released DNA is purified by the magnetic bead method. The surface of the magnetic beads is modified with specific functional groups, which can specifically bind to DNA under specific conditions. Through magnetic separation technology, impurities in the lysis solution can be effectively removed, thereby obtaining a high-purity DNA sample. The purified DNA is introduced into a microfluidic chip pre-loaded with freeze-dried reagents, and a multiplex LAMP (loop-mediated isothermal amplification) amplification reaction is carried out using the primer sets shown in SEQ ID NO: 1 - 6. The reaction process needs to be strictly controlled at a constant temperature of 62 - 65 °C for 25 - 35 minutes. During the amplification process, the genus and pathogenicity of the pathogen are determined by real-time monitoring of the fluorescence signal. The LAMP amplification technology has the characteristics of high efficiency and strong specificity, and can rapidly amplify the target DNA sequence under isothermal conditions. The intensity of the fluorescence signal generated is proportional to the amount of amplification product. By comparing with a preset standard curve, it can be accurately judged whether there is Fusarium fujikuroi in the sample and the specific type of the pathogen.

[0043] Synchronously, a hyperspectral imager is used to collect images of the surface of rice seeds. The hyperspectral imager can obtain image information of the seeds under multiple consecutive spectral bands, covering rich spectral features. The collected images are input into a pre-trained dual-branch CNN model. The first branch of this model focuses on extracting the spatial texture features of the seed images, and layer-by-layer analyzes and abstracts the texture details of the images through a series of convolutional layers. The second branch is mainly responsible for extracting spectral sequence features and exploring the internal connections between different spectral bands. After the operation and processing of the model, the results of the pathogen infection probability and pathogen classification are finally output. With its powerful feature learning ability, the dual-branch CNN model can effectively identify the subtle features related to the infection of bakanae disease in the seed images, providing an accurate spectral analysis basis for disease detection. Based on the dynamic weight fusion algorithm, the results of molecular detection and spectral analysis are integrated. In this algorithm, the molecular detection weight is set to 60% ± 5%, and the spectral analysis weight is 40% ± 5%. According to the confidence levels and respective advantages of the two detection methods, the weight ratio is dynamically adjusted to achieve more accurate result fusion. Finally, a comprehensive report is generated, and the report content includes pathogen type, pathogenic type, infection index, and confidence level. This comprehensive report provides comprehensive, accurate, and highly credible information for the detection of bakanae disease in rice seeds, helping relevant personnel to take effective prevention and control measures in a timely manner.

[0044] As Figure 2 shown, the dynamic weight fusion algorithm integrates the results of molecular detection and spectral analysis based on weighted summation. Let the confidence level of molecular detection be C1, the confidence of spectral analysis be C2, the confidence level weight of molecular detection be W1, whose value range is between 55% and 65%, and the spectral analysis weight be W2, whose value range is between 35% and 45%, and W1 + W2 = 1. The calculation formula for the detection confidence level is:

[0045] C = C1 × W1 + C2 × W2

[0046] Using the dynamic weight fusion algorithm for the results of molecular detection and spectral analysis, when the value range of the molecular detection confidence level C1 is 0.8 and the spectral analysis confidence level C2 is 0.6, according to the detection scenario, the molecular detection weight W1 = 0.6 is set, then the spectral analysis weight W2 = 1 - 0.6 = 0.4. After substituting the values into C = C1 × W1 + C2 × W2, the detection confidence level C = 0.8 × 0.6 + 0.6 × 0.4 = 0.48 + 0.24 = 0.72 can be obtained. In different detection scenarios, the reliability and efficiency of the two detection methods are different, so the weights need to be dynamically adjusted.

[0047] As Figure 3 shown, the detection system for bakanae disease types in rice seeds includes the following modules:

[0048] S1, nucleic acid extraction module: integrated lysis solution storage tank, microcentrifuge (speed 12000±500rpm) and magnetic bead purification component (magnetic bead particle size 0.1-1μm), supporting single sample processing time ≤15 minutes;

[0049] S2, constant temperature amplification module: using a microfluidic chip (size ≤5cm×3cm×0.5cm) and a semiconductor temperature control unit (accuracy ±0.3℃), the chip contains three independent reaction chambers, which are pre-loaded with freeze-dried primers targeting the RPB2 gene (genus specificity), FfPS1 gene (gibberellin type) and FUM1 gene (toxin type);

[0050] S3, Hyperspectral imaging module: includes near-infrared hyperspectral camera (wavelength range 900-1700nm, spatial resolution 0.2mm / pixel), LED ring light source (color temperature 5000K±200K) and autofocus mechanism, supporting single seed imaging time ≤3 seconds;

[0051] S4, data analysis module:

[0052] (1) Embedded AI processor, running a dual-branch CNN model (model size ≤ 500MB);

[0053] (2) Dynamic weight fusion algorithm;

[0054] (3) User interface (touch screen, supports Chinese / English / Japanese / Hindi), which can generate ISO17025 standard certification report;

[0055] S5, communication module: supports 4G / 5G / Wi-Fi wireless connection, and synchronizes data to the cloud pathogen database in real time (including spectra and genetic characteristics of ≥1000 Fusarium strains).

[0056] The nucleic acid extraction module integrates multiple key components to ensure a fast and high-quality nucleic acid extraction process. Inside the module, there is a lysis buffer storage tank that can stably store a sufficient amount of composite lysis buffer, providing the necessary reagent for the lysis of seed samples. The paired micro centrifuge has excellent performance, with its rotation speed precisely controlled at 12,000 ± 500 rpm, capable of generating a powerful centrifugal force in a short time, promoting the full fragmentation of cells in the sample and effectively separating biological macromolecules such as nucleic acids. The magnetic beads used have an ideal particle size range of 0.1 - 1 μm. This particle size gives the magnetic beads a large specific surface area, enabling them to efficiently adsorb nucleic acid molecules, while also ensuring good dispersibility and fluidity in the solution system. With this optimized combination of components, the entire nucleic acid extraction module supports a single-sample processing time of ≤ 15 minutes, greatly improving the detection efficiency and meeting the requirements of rapid detection. Inside the chip, three independent reaction chambers are carefully designed, and each reaction chamber is pre-loaded with lyophilized primers targeting different genes. Among them, one reaction chamber targets the RPB2 gene, which has genus specificity and can be used to preliminarily determine the genus to which the pathogen belongs; another reaction chamber is pre-loaded with primers targeting the FfPS1 gene (gibberellin type) for detecting the pathogenic type related to gibberellin synthesis; the third reaction chamber is pre-loaded with primers targeting the FUM1 gene (toxin type) to determine whether the pathogen has the ability to produce toxins. Together with the semiconductor temperature control unit, whose temperature control accuracy can reach ±0.3 °C, it can provide a highly stable constant-temperature environment for the LAMP amplification reaction, ensuring the efficient and accurate progress of the amplification reaction.

[0057] The data analysis module is centered around an embedded AI processor and runs a pre-trained dual-branch CNN model. This model has been trained and optimized with a large amount of data, with a model size of ≤ 500 MB. While ensuring efficient operation, it has powerful image recognition and data analysis capabilities. The design of the dual-branch CNN model aims to fully exploit the spatial texture features and spectral sequence features in hyperspectral images. Through the collaborative work of the two branches, it can accurately determine whether rice seeds are infected by bakanae disease-causing bacteria and accurately predict the type and degree of infection of the bacteria. Based on the dynamic weight fusion algorithm, it gives play to the advantages of different detection methods, effectively making up for the limitations of single detection methods. Subsequently, the user interface adopts a touch screen design, which is convenient, intuitive, and supports multiple languages such as Chinese, English, and Japanese, facilitating users from different regions. Then, the wireless communication module is used to synchronize the detection data to the cloud pathogen database in real time. This database is rich in resources, containing the spectral and gene characteristics of ≥ 1000 Fusarium strains, providing a solid data foundation for the comparative analysis of detection results and further research.

[0058] The above front, back, left, right, up, and down are all based on the Figure 1Based on [a certain reference], with the perspective of the person's observation as the standard, the side of the device facing the observer is defined as the front, the left side of the observer is defined as the left, and so on.

[0059] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "center", "longitudinal", "lateral", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation on the protection scope of the present invention.

[0060] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting the type of Fusarium moniliforme in rice seeds, characterized in that: It includes the following steps: A1. Mix the rice seeds to be tested with a composite lysis solution. The lysis solution contains 0.5% - 1.5% CTAB, 0.1 - 0.3 mg / mL proteinase K, 1% - 3% polyvinylpyrrolidone (PVP), and 0.05% - 0.2% mercaptoethanol. Shake and lyse at 60 - 70 °C for 10 - 20 minutes, and purify the DNA by the magnetic bead method; A2. Import the purified DNA into a microfluidic chip pre-loaded with freeze-dried reagents, and perform multiplex LAMP amplification using the primer sets shown in SEQ ID NO: 1 - 6. The reaction condition is a constant temperature of 62 - 65 °C for 25 - 35 minutes, and determine the pathogen genus and pathogenicity through real-time fluorescence signals; A3. Synchronously collect the seed surface images using a hyperspectral imager and input them into a pre-trained dual-branch CNN model. The first branch extracts the convolutional layer of spatial texture features, and the second branch extracts the convolutional layer of spectral sequence features, and outputs the pathogen infection probability and the strain classification result; A4. Integrate the results of step A1 and step A3 based on the dynamic weight fusion algorithm to generate a comprehensive report including pathogen type, pathogenicity, infection index (0% - 100%), and confidence level (≥90%); 2. The method for detecting the type of Fusarium moniliforme in rice seeds according to claim 1, wherein: The fluorescence signal detection threshold for the LAMP amplification is three times the standard deviation of the baseline fluorescence value, and the positive determination needs to meet two target genes. The training data of the dual-branch CNN model includes hyperspectral images of healthy seeds, early-infected seeds (asymptomatic), and late-infected seeds, and the data augmentation techniques of random illumination transformation and noise injection are used during training; 3. The method for detecting the type of bakanae disease bacteria in rice seeds according to claim 2, characterized in that: In the dynamic weight fusion algorithm, if the confidence level difference between the molecular detection and the spectral analysis results is ≥20%, an artificial review prompt is triggered. The hyperspectral imaging module supports batch scanning of multiple seeds and automatically locates the single-seed area through an image segmentation algorithm; 4. The method for detecting the type of bakanae disease bacteria in rice seeds according to claim 3, characterized in that: In the dynamic weight Bayesian network, the initial weight of the molecular detection result is 55% - 65%, and the initial weight of the spectral analysis result is 35% - 45%. The weight ratio is dynamically adjusted according to the model confidence level; 5. The method for detecting the type of bakanae disease bacteria in rice seeds according to claim 4, characterized in that: The surface of the microfluidic chip is coated with a trehalose-bovine serum albumin (BSA) composite protective layer and can be stored at 25 °C for more than 6 months. The embedded control unit supports the switching between the offline mode and the online mode. In the online mode, it can be connected to the cloud server through 5G / Wi-Fi to update the pathogen feature database in real time; 6. A method for detecting the type of Fusarium moniliforme in rice seeds according to claim 5, characterized in that: The dynamic weight fusion algorithm integrates the molecular detection and spectral analysis results based on the weighted summation method. Let the confidence level of the molecular detection be C1, the confidence level of the spectral analysis be C2, the confidence level weight of the molecular detection be W1, and its value range is between 55% - 65%. The spectral analysis weight is W2, and its value range is between 35% - 45%. And it satisfies W1 + W2 = 1. The calculation formula for the detection confidence level is: C = C1×W1 + C2×W2.

7. A method for detecting the type of bakanae disease bacteria in rice seeds according to claim 6, characterized in that: The image segmentation algorithm starts from a certain seed pixel point or area in the image and gradually grows or merges into a complete seed area according to a certain similarity criterion.

8. The method for detecting the type of Fusarium moniliforme in rice seeds according to claim 7, characterized in that: The embedded control unit supports users to upload private data to the cloud pathogen database and updates the global model parameters through federated learning technology.

9. A detection system for the type of bakanae disease pathogen in rice seeds, which applies the detection method for the type of bakanae disease pathogen in rice seeds as described in claim 8, is characterized in that, The system includes: S1. Nucleic acid extraction module: Integrating a lysis buffer storage tank, a micro centrifuge (rotation speed 12000 ± 500 rpm), and a magnetic bead purification component (magnetic bead particle size 0.1 - 1 μm), supporting a single sample processing time ≤ 15 minutes; S2. Isothermal amplification module: Adopting a microfluidic chip (size ≤ 5 cm × 3 cm × 0.5 cm) and a semiconductor temperature control unit (accuracy ± 0.3 °C), with three independent reaction chambers in the chip, pre-loaded with lyophilized primers targeting the RPB2 gene (specificity), FfPS1 gene (gibberellin type), and FUM1 gene (toxin type) respectively; S3. Hyperspectral imaging module: Comprising a near-infrared hyperspectral camera (wavelength range 900 - 1700 nm, spatial resolution 0.2 mm / pixel), an LED ring light source (color temperature 5000K ± 200K), and an auto-focus mechanism, supporting a single seed imaging time ≤ 3 seconds; S4. Data analysis module: (1) An embedded AI processor running a dual-branch CNN model (model size ≤ 500 MB); (2) A dynamic weight fusion algorithm; (3) A user interface (touch screen, supporting Chinese / English / Japanese / Indian languages), capable of generating ISO17025 standard certification reports; S5. Communication module: Supporting 4G / 5G / Wi-Fi wireless connections, and synchronizing data to the cloud pathogen database (including spectra and gene characteristics of ≥ 1000 Fusarium strains) in real time.

Images