A smartphone fluorescence data acquisition device and a method for detecting potato dry rot disease

Abstract

A smartphone fluorescence data acquisition device and a method for detecting potato dry rot belong to the technical field of crop pest identification. In order to solve the problem of rapid identification and detection of potato dry rot, the oblique incident lighting module of the present invention includes an excitation light source, a first lens, a first reflector, a second reflector, and a second lens; the intelligent control module includes an intelligent controller, a first stepper motor, and a second stepper motor; the fluorescence detection module includes a stage, a triplet lens, a filter, and a mobile phone camera; the LED excitation light emitted by the excitation light source is irradiated to the first reflector through the first lens, then reflected to the second reflector, and then incident on the sample on the stage through the second lens, and then the fluorescence signal emitted by the sample is collected through the triplet lens, and the sample fluorescence signal light is obtained through the filter, and the mobile phone camera shoots the fluorescence image and transmits it to the data storage module. The present invention realizes more sensitive, faster, and smarter nucleic acid detection of potato dry rot.

Description

Technical Field

[0001] The invention belongs to the technical field of crop disease and insect pest identification, and particularly relates to a smart phone fluorescence data acquisition device and a method for detecting potato dry rot thereof. Background Art

[0004] The biological detection method of potato dry rot is mainly based on the characteristics of the pathogen, which is slow and requires high professional knowledge. Molecular biological methods are widely used, such as polymerase chain reaction (PCR). This detection technology has high sensitivity, but the cost is also high and the reaction conditions are relatively harsh. Early diagnosis and detection of pathogens are the main way to control the occurrence of diseases, and are also the most economical, healthy and effective measures. However, due to the lack of obvious symptoms in the early stage of infection, it is difficult to judge by naked eye observation, and the concentration of pathogens is low, so some detection methods are difficult to detect results. With the development of agricultural technology, there is an urgent need for more sensitive, faster, cheaper, smarter and easier-to-operate detection technologies to serve production, control the occurrence of soil-borne diseases, and reduce the production cost of potato seed potatoes. Therefore, it is particularly important to establish a technical system for accurate early identification and trace rapid detection of potato dry rot, which is of great strategic significance to ensure the healthy development of the potato industry. Summary of the invention

[0005] The problem to be solved by the present invention is to quickly identify and detect potato dry rot, and a smart phone fluorescence data acquisition device and a method for detecting potato dry rot are proposed.

[0006] To achieve the above object, the present invention is implemented through the following technical solutions:

[0007] A fluorescence data acquisition device for a smart phone, comprising an oblique-incident illumination module, an intelligent control module, a fluorescence detection module, and a data storage module;

[0008] The oblique-incidence illumination module includes an excitation light source, a first lens, a first reflector, a second reflector, and a second lens; the intelligent control module includes an intelligent controller, a first stepper motor, and a second stepper motor; the fluorescence detection module includes a stage, a triplet lens, a filter, and a mobile phone camera;

[0009] The LED excitation light emitted by the excitation light source is irradiated to the first reflector through the first lens, then reflected to the second reflector, and then incident on the sample on the stage through the second lens, and then the fluorescence signal emitted by the sample is collected through the triplet lens, and the sample fluorescence signal light is obtained through the filter, and the mobile phone camera takes a fluorescence image and transmits it to the data storage module;

[0010] The intelligent controller is connected to the first stepper motor and the second stepper motor respectively, the first stepper motor is connected to the stage, and the second stepper motor is connected to the filter.

[0011] Furthermore, the first reflector and the second reflector are neutral attenuation reflectors, the first lens is an aspheric lens, and the second lens is a broadband achromatic lens; the excitation light source is an LED excitation light source with a wavelength of 532nm; and the obtained alignment light is irradiated onto the sample through the second lens in a 75° illumination manner.

[0012] Furthermore, the pixel of the mobile phone camera is 20Mp.

[0013] Furthermore, the first stepper motor is used to precisely control the stage with an accuracy of micrometers, and the second stepper motor is used to adjust the filter to obtain a clear image of the sample with a clear focal plane. The two motors are synchronously controlled by external triggering, and the operation of the motors is controlled by Arduino software.

[0014] A method for detecting potato dry rot is implemented based on the above-mentioned smartphone fluorescence data collection device, and comprises the following steps:

[0015] S1. Amplify the target nucleic acid and protein of potato based on RPA type III-A CRIPSR multi-stage signal amplification detection system to prepare potato dry rot detection samples;

[0016] S2. placing a potato dry rot test sample on a stage of a smartphone fluorescence data acquisition device, starting the smartphone fluorescence data acquisition device, and collecting a fluorescence image of the potato dry rot test sample;

[0017] S3. Inputting the potato dry rot fluorescence image obtained in step S2 into the trained neural network model for feature recognition, outputting the feature as the detected Fusariales nucleic acid concentration, and obtaining the potato dry rot detection result;

[0018] S4. For the potato samples detected to have potato dry rot in step S3, potato tuber images are collected and sent to a U-Net-based segmentation model to determine the severity level of the potato disease.

[0019] Furthermore, the specific implementation method of step S1 includes extracting DNA from potatoes, amplifying the target DNA fragment by RPA, and re-transcription to obtain an RNA test sample, and then reacting with 250 μM ATP, 500 nM fluorescent reporter RNA, 500 nM TtCsm complex and 2500 nM TtCsm6 in a reaction buffer with a 30 μL reaction system, and incubating the reaction at 60°C to obtain a potato dry rot test sample, the reaction buffer including 20 mM Tris-HCl with a pH of 7.9, 200 mM monopotassium glutamate, 10 mM ammonium sulfate, 5 mM magnesium sulfate, and 1 mM TCEP.

[0020] Furthermore, the trained neural network model used in step S3 is a Transformer neural network, and the specific training process includes the following steps:

[0021] S3.1. Collect 200 sets of fluorescence images of each concentration gradient, remove the noise of the negative control group and each concentration group by limiting the contrast adaptive histogram equalization method, and then divide them into training set, validation set and test set according to the ratio of 3:1:1; for the training set, perform data enhancement on the original image data by random flipping, mirroring and Gaussian smoothing methods, and finally convert all images into grayscale images;

[0022] S3.2. The neural network model is constructed as a Transformer neural network, which is mainly composed of an input part, a multi-layer encoder, a multi-layer decoder, and an output part. The setting includes N layers of encoders and N layers of decoders, and N in Transformer is 6;

[0023] Set each encoder layer to have two sub-layers, namely the Multi-Head Attention layer and the Position-wise Feed-Forward Network, with a residual connection and a layer normalization LayerNorm operation behind each sub-layer;

[0024] Each decoder layer has three sub-layers, Masked Self-Attention layer, Encoder-Decoder attention layer, and position-by-position feedforward neural network, with residual connection and layer normalization LayerNorm operation after each sub-layer;

[0025] The attention mechanism calculates the attention score by matching the query tensor Q with the key tensor K, converts the score into the weighted value tensor V, and obtains the final attention vector. The attention vector calculation formula is as follows:

[0026]

[0027] Among them, softmax is the attention weight, is the scaling factor, d k is the dimension size of Q or K;

[0028] In the multi-head attention mechanism, h attention heads run in parallel, and each attention head has an attention query matrix, a key matrix, and a value matrix. And output, then concatenate the outputs of all attention heads, and pass through the weight matrix W o Complete the linear transformation and get the final output. The calculation formula is as follows:

[0029] MultiHead(Q,K,V)=Concat(head1,…,head h )W o

[0030]

[0031] S3.3. Input the training set obtained in step S3.1 into the Transformer neural network constructed in step S3.2. The label of the data set is the initial concentration of Fusarium before amplification. Observe whether the network output results are referenceable. In this process, continuously optimize the parameters through back propagation to obtain the optimal operation parameters and predicted values. In the model improvement part, observe the evaluation effect of the test set on the training model. According to the evaluation results, continuously optimize the network depth and related parameter settings to obtain a trained neural network.

[0032] Furthermore, the U-Net-based segmentation model used in step S4 classifies all pixels in the image into three categories: lesions, healthy tubers, and background. The total number of categories is defined as k, k = 3. The confusion matrix p is introduced to represent the classification results of all pixels in the tuber image, and p is used to represent the classification results of all pixels in the tuber image. ij Represents pixels whose true category is i and predicted by the model as j. The following three evaluation indicators are used to evaluate the performance of the segmentation model:

[0033] The average intersection over union (mIoU) is used to measure the overlap between the pixels in the real area and the pixels in the predicted area. The calculation method is shown in the following formula:

[0034]

[0035] The set similarity measurement function Dice is used: an indicator to measure the similarity between the segmented pixel samples predicted by the model and the real pixel samples. The calculation method is shown in the following formula:

[0036]

[0037] The global accuracy GC is used to calculate the ratio of all correctly predicted pixels to all pixels in a tuber image. The calculation method is as follows:

[0038]

[0039] Furthermore, when the U-Net-based segmentation model is used for grading in step S4, the tuber image is input into the trained U-Net-based segmentation model, the area ratio of the lesions is calculated according to the segmentation result, and the potato dry rot lesion grading standard expressed in symbolic form is integrated into the model. The model finally performs severity grading based on the lesion area ratio and outputs it.

[0040] Beneficial effects of the present invention:

[0041] A method for detecting potato dry rot described in the present invention realizes the detection of potato dry rot by combining a III-A type CRISPR system based on RPA, a portable smartphone fluorescence microscope, an artificial intelligence potato dry rot nucleic acid, and an image detection method. The CRISPR / Cas system is a gene editing technology that recognizes and binds to a specific RNA or DNA sequence after the Cas protein binds to the corresponding crRNA, and cuts or modifies the target DNA or RNA sequence. The type III CRISPR / Cas system triggers a cascade reaction of signals after specifically recognizing the target RNA. In the CRISPR / Cas10 system, the binding of the Csm complex to RNA triggers conformational changes, thereby activating the Palm domain of the Cas10 subunit, which converts ATP into approximately 1,000 cyclic oligoadenylic acids (cOA, such as cA 4). The binding of cOA to csm6 stimulates the RNase activity of csm6, which can cut single-stranded RNA of random sequence, including cutting fluorescent reporter RNA. After the fluorescent reporter RNA is cut, the quenching group moves away, and the fluorescent group emits fluorescence. Fluorescence microscopy can capture fluorescent images before and after the CRISPR system is activated. The deep learning model trained with fluorescence images corresponding to Fusarium at different concentration gradients can directly predict the concentration of Fusarium in the sample through the fluorescence image, that is, to detect potato dry rot. If the cross-section image of the potato tuber can be obtained, the potato tuber image can be sent to the U-Net-based segmentation model to calculate the area ratio of the dry rot lesions. At the same time, the grading standard represented by symbols is integrated into the model, and the severity of the disease is determined by comparing it with the area ratio represented by the same symbol.

[0042] The method for detecting potato dry rot described in the present invention realizes more sensitive, faster and smarter potato dry rot nucleic acid detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1This is a schematic structural diagram of a fluorescence data acquisition device for a smartphone according to the present invention;

[0044] Figure 2 A flow chart of a method for detecting potato dry rot according to the present invention;

[0045] Figure 3 This is a schematic diagram of a process for detecting potato dry rot according to the present invention;

[0046] Figure 4 A flowchart of the multi-stage signal amplification nucleic acid detection process of the RPA-based type III-A CRISPR system;

[0047] Figure 5 This is a flow chart for determining the severity level of potato disease in the U-Net segmentation model of the present invention;

[0048] Figure 6 Schematic diagram of the segmentation model structure of U-Net of the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the present invention more clear, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the specific embodiments described are only part of the embodiments of the present invention, rather than all of the specific embodiments. The components of the specific embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations, and the present invention can also have other embodiments.

[0050] Therefore, the following detailed description of the specific embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected specific embodiments of the present invention. Based on the specific embodiments of the present invention, all other specific embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0051] In order to further understand the content, features and effects of the present invention, the following specific implementation methods are given as examples, and the attached Figure 1 -Attached Figure 6 The detailed instructions are as follows:

[0052] Embodiment 1:

[0053] A fluorescence data acquisition device for a smart phone, comprising an oblique incident illumination module 1, an intelligent control module 2, a fluorescence detection module 3, and a data storage module 4;

[0054] The oblique-incidence illumination module 1 includes an excitation light source 1-1, a first lens 1-2, a first reflector 1-3, a second reflector 1-4, and a second lens 1-5; the intelligent control module 2 includes an intelligent controller 2-1, a first stepper motor 2-2, and a second stepper motor 2-3; the fluorescence detection module 3 includes a stage 3-1, a triplet lens 3-2, a filter 3-3, and a mobile phone camera 3-4;

[0055] The LED excitation light emitted by the excitation light source 1-1 is irradiated to the first reflector 1-3 through the first lens 1-2, then reflected to the second reflector 1-4, and then incident on the sample on the stage 3-1 through the second lens 1-5, and then the fluorescence signal emitted by the sample is collected by the triplet lens 3-2, and the sample fluorescence signal light is obtained through the filter 3-3, and the mobile phone camera 3-4 takes a fluorescence image and transmits it to the data storage module 4;

[0056] The intelligent controller 2-1 is connected to a first stepper motor 2-2 and a second stepper motor 2-3 respectively, the first stepper motor 2-2 is connected to a stage 3-1, and the second stepper motor 2-3 is connected to a filter 3-3.

[0057] Furthermore, the first reflector 1-3 and the second reflector 1-4 are neutral attenuation reflectors, the first lens 1-2 is an aspheric lens, and the second lens 1-5 is a broadband achromatic lens; the excitation light source 1-1 is an LED excitation light source with a wavelength of 532nm; the obtained alignment light is irradiated onto the sample through the second lens 1-5 in a 75° illumination manner.

[0058] Furthermore, the pixel of the mobile phone camera 3-4 is 20Mp.

[0059] Furthermore, the first stepper motor 2-2 is used to precisely control the stage 3-1 with an accuracy of micrometers; the second stepper motor 2-3 is used to adjust the filter 3-3 to obtain a clear image of the sample captured in a clear focal plane, and the two motors are synchronously controlled by external triggering, and the operation of the motors is controlled by Arduino software.

[0060] Embodiment 2:

[0061] A method for detecting potato dry rot is implemented based on the smartphone fluorescence data acquisition device described in Example 1, comprising the following steps:

[0062] S1. Amplify the target nucleic acid and protein of potato based on RPA type III-A CRIPSR multi-stage signal amplification detection system to prepare potato dry rot detection samples;

[0063] Further, the specific implementation method of step S1 includes extracting DNA from potatoes, amplifying the target DNA fragment by RPA, and re-transcription to obtain an RNA test sample, and then reacting with 250 μM ATP, 500 nM fluorescent reporter RNA, 500 nM TtCsm complex and 2500 nM TtCsm6 in a reaction buffer with a reaction system of 30 μL, and incubating the reaction at 60° C. to obtain a potato dry rot test sample, wherein the reaction buffer includes 20 mM Tris-HCl with a pH of 7.9, 200 mM monopotassium glutamate, 10 mM ammonium sulfate, 5 mM magnesium sulfate, and 1 mM TCEP;

[0064] Furthermore, the key enzyme components of the RPA reaction system, including recombinase, single-strand binding protein, and strand displacement DNA polymerase, were used to amplify the specific nucleic acid fragment of the target pathogen, Fusarium. First, Escherichia coli expression cells were established through relevant plasmids to express the key protease of RPA, and the expression amount was increased by optimizing relevant parameters. The expressed protein was then purified and extracted by rapid liquid phase mass spectrometry. Secondly, RPA primers were designed according to the target nucleic acid fragment of Fusarium, and the candidate primers were tested and screened. The primer screening of RPA is mainly divided into the following steps: selecting the target region, designing candidate primers, screening candidate primers, and improving performance and screening again. Finally, the binding and amplification functions of proteins and nucleic acids were used to observe the size and amplification amount of nucleic acid bands after amplification to verify the protein activity and the practicality of the entire RPA system.

[0065] Furthermore, the type III-A CRISPR system includes protein subtypes Csm1-6, so the corresponding plasmids and Escherichia coli are also used to express related proteins and purify them. The crRNA targeting Fusarium nucleic acid and the non-specifically sheared fluorescent reporter single-stranded RNA are designed and synthesized. The 3' end of the fluorescent reporter RNA is labeled with the quencher group BHQ1, and the 5' end is labeled with the fluorescent group 6-FAM. The Fusarium target RNA fragment is obtained by in vitro transcription after RPA amplification, and the cascade amplification function induced by the type III-A CRISPR protein complex after recognizing the target nucleic acid is verified to generate a fluorescent signal that can be used for detection. Through the independently developed mobile phone fluorescence detection system, the fluorescent signal emitted by the fluorescent group due to the distance of the quencher group after shearing can be observed, such as Figure 4 shown.

[0066] Furthermore, the genomic database is used to compare and obtain the nucleic acid sequence of the Fusarium-specific gene. The nucleic acid sequence of these specific genes is obtained through the synthesis, amplification and in vitro transcription of the plasmid, and the accuracy of the nucleic acid sequence is ensured by sequencing. The experimental waste will be treated with protease and nuclease to degrade the protein and nucleic acid in the waste liquid, and finally sterilized at high temperature and high pressure for professional recycling.

[0067] Furthermore, the target DNA fragment is amplified in vitro. First, the key RPA protease is expressed and purified. The nucleic acid sequence encoding the protein is inserted into the plasmid pET30 to construct a protein high expression vector, and a His-tag is designed at the 5' end, and subsequent protein purification can be performed based on the protein tag and its isoelectric point. After sequencing verification, the expression plasmid is transformed into E.coli BL21 competent cells and spread on a resistant agar plate for overnight culture. Transfer from solid culture medium to liquid culture, add a final concentration of 0.1-0.5mM isopropyl-beta-D-thiogalactopyranoside (IPTG) for full induction, and culture at 16-20℃ for 18-20h and then collect the bacteria by centrifugation. The protein is separated from the bacteria by ultrasonic disruption for subsequent purification. The affinity chromatography between His-Tag and nickel ions is used to wash away the impurities that bind non-specifically to the column, and finally the eluate containing the target protein is eluted, which is further concentrated by dialysis to obtain a high concentration of the target protein. An equal volume of glycerol is added to the concentrated and collected sample, and the sample concentration is determined by the Coomassie Brilliant Blue method. The protein size is then checked by SDS-PAGE electrophoresis and the purity is analyzed by Image J. The purified protein is stored in a -80°C refrigerator after passing through liquid nitrogen. Among them, the IPTG induction concentration, culture temperature, culture time, eluent imidazole concentration and other conditions need to be designed according to different proteins. Gradient experiments to obtain the best protein expression and purification conditions.

[0068] Further, the RPA primers required for amplifying the target DNA fragment are designed. The RPA reaction can quickly amplify DNA fragments of 100-300bp. When selecting the target sequence, the following are followed: i) the GC content is between 40% and 60%; ii) the forward / reverse repeats and palindromic sequences are minimized; iii) the target sequence is highly conserved within the same fungal / viral genus. When designing primers, the following are followed: i) the RPA primer length is 30-35 nucleotides. Too long will cause the primer to easily form a secondary structure, and too short will lead to low amplification efficiency; ii) the primer GC% is 20-70%; iii) the primer Tm is 50-100℃; iv) the maximum allowable single nucleotide repeat length in the primer is 5. Slight changes in the primer oligonucleotide sequence sometimes lead to significant changes in primer activity, so multiple pairs of upstream and downstream primers should be designed for pairing and screening. The RPA reaction system includes vortexing after adding the above substances and incubating at 39℃ for 30 minutes. In addition, since the target nucleic acid to be detected is RNA, the T7 promoter sequence required for in vitro transcription needs to be added to the primer for subsequent in vitro transcription.

[0069] Further, RPA key protease, RPA primer and target DNA fragment are added to the reaction buffer, and the amplification reaction is carried out at room temperature or 37°C. The recombinase first forms a complex with the primer. Under the mediation of the primer, the complex scans the DNA homologous sequence for positioning, triggering a strand exchange reaction to form a D-Loop structure. In this step, the concentration ratio of the recombinase and the primer will affect the efficiency of the complex formation. Subsequently, with the help of the single-stranded binding protein, the recombinase binds to the primer and the target DNA to form a stable structure and then dissociates from the primer. The 3' end of the primer is exposed and recognized by the DNA polymerase, and the DNA amplification process is started. During the amplification process, the concentration of enzyme, ATP, creatine phosphate, creatine kinase, the ratio of primer to template and the reaction time will all affect the efficiency of DNA amplification. According to the existing RPA reaction system, the optimal amplification reaction conditions are explored and optimized. The DNA amplified by RPA is transcribed in vitro to obtain the target RNA fragment of Fusarium.

[0070] Furthermore, Escherichia coli was used to express the related proteins of the III-A type CRISPR system. First, 1) a mutant (TtCsm3-D34A) expression vector of TtCsm1-5 carrying a His tag was constructed; 2) an expression vector of the auxiliary protein TtCsm6 carrying a His tag, and 3) a wild-type expression vector of the subunit protein TtCsm1-5 carrying a SNAP-tag tag were constructed, and the constructed vectors were sequenced to ensure the accuracy of the sequence of the encoded protein.

[0071] To express the pre-crRNA sequence, the relevant plasmids and nucleic acids were designed based on the CDS fragment of the translation elongation factor 1-α gene of the dry rot pathogen Fusarium solani. The sequence is as follows:

[0072] crRNA1 spacer(40nt):GATGGTTCGCTTATCGATACCACCGCACTGGTAGATCAAG

[0073] crRNA2 spacer (40nt): CACGCTCACGCTCGGCCTTGAGCTTGTCAAGAACCCAGGC

[0074] crRNA3 spacer(40nt):CAATCGGCCTGGGAAGTACCAGTGATCATGTTCTTGATGA

[0075] crRNA4 spacer(40nt):CAATCGGCCTGGGAAGTACCAGTGATCATGTTCTTGATGA

[0076] Fluorescent reporter RNA: / 56-FAM / UUUUU / 3BHQ1

[0077] Plasmid design:

[0078] The 4 spacers in plasmid #127764pre-crRNA (bases 75-414) were replaced with the designed crRNA spacer described above.

[0079] Among them, the fluorescent reporter RNA was synthesized by Beijing Qingke Biotechnology Co., Ltd.; the plasmid was prepared by Yunzhou Biotechnology (Guangzhou) Co., Ltd.

[0080] The protein was induced to express, purified and stored.

[0081] S2. placing a potato dry rot test sample on a stage of a smartphone fluorescence data acquisition device, starting the smartphone fluorescence data acquisition device, and collecting a fluorescence image of the potato dry rot test sample;

[0082] Furthermore, a suitable filter is selected according to the wavelength of the excitation light source, so that an intermediate image (immediate image) is formed through a broadband triplet lens, and finally a fluorescence signal image is captured through the mobile phone lens, and a CMOS camera is used for signal detection and an optimized fluorescence image is output through the built-in ISP system. The triplet lens is used to eliminate the chromatic aberration generated during the imaging process, and the filter eliminates the noise generated by the light source, leaving only the sample fluorescence signal light, thereby improving the signal-to-noise ratio;

[0083] Furthermore, the portable smartphone fluorescence microscope consists of a 3D-printed shell and optical components as well as a smartphone. The imaging and analysis process of the sample is as follows: the sample is placed on the stage and inserted into the device. The LED excitation light is illuminated by two mirrors at a 75° oblique incidence, and the broadband achromatic lens focuses the light onto the sample. The fluorescence signal emitted by the sample is then collected by a triplet lens to form an intermediate image in front of the phone lens. Finally, the smartphone is used to detect and capture the fluorescence signal image, store it and display it.

[0084] Furthermore, the purified protein, fluorescent reporter RNA, etc. were mixed, and the microscope system was used to capture images before and after the addition of the sample, and 200 sets of fluorescent images for each concentration gradient were collected.

[0085] S3. Inputting the potato dry rot fluorescence image obtained in step S2 into the trained neural network model for feature recognition, outputting the feature as the detected Fusariales nucleic acid concentration, and obtaining the potato dry rot detection result;

[0086] Furthermore, the trained neural network model used in step S3 is a Transformer neural network, and the specific training process includes the following steps:

[0087] S3.1. Collect 200 sets of fluorescence images of each concentration gradient, remove the noise of the negative control group and each concentration group by limiting the contrast adaptive histogram equalization method, and then divide them into training set, validation set and test set according to the ratio of 3:1:1; for the training set, perform data enhancement on the original image data by random flipping, mirroring and Gaussian smoothing methods, and finally convert all images into grayscale images;

[0088] S3.2. The neural network model is constructed as a Transformer neural network, which is mainly composed of an input part, a multi-layer encoder, a multi-layer decoder, and an output part. The setting includes N layers of encoders and N layers of decoders, and N in Transformer is 6;

[0089] Set each encoder layer to have two sub-layers, namely the Multi-Head Attention layer and the Position-wise Feed-Forward Network, with a residual connection and a layer normalization LayerNorm operation behind each sub-layer;

[0090] Each decoder layer has three sub-layers, Masked Self-Attention layer, Encoder-Decoder attention layer, and position-by-position feedforward neural network, with residual connection and layer normalization LayerNorm operation after each sub-layer;

[0091] The attention mechanism calculates the attention score by matching the query tensor Q with the key tensor K, converts the score into the weighted value tensor V, and obtains the final attention vector. The attention vector calculation formula is as follows:

[0092]

[0093] Among them, softmax is the attention weight, is the scaling factor, d k is the dimension size of Q or K;

[0094] In the multi-head attention mechanism, h attention heads run in parallel, and each attention head has an attention query matrix, a key matrix, and a value matrix. And output, then concatenate the outputs of all attention heads, and pass through the weight matrix W o Complete the linear transformation and get the final output. The calculation formula is as follows:

[0095] MultiHead(Q,K,V)=Concat(head1,…,head h )W o

[0096]

[0097] S3.3. Input the training set obtained in step S3.1 into the Transformer neural network constructed in step S3.2. The label of the data set is the initial concentration of Fusarium before amplification. Observe whether the network output result is referenceable. In this process, continuously optimize the parameters through back propagation to obtain the best operation parameters and predicted values. In the model improvement part, observe the evaluation effect of the test set on the training model. According to the evaluation results, continuously optimize the network depth and related parameter settings to obtain a trained neural network.

[0098] Furthermore, a Transformer-based neural network was built using Python, and the images and corresponding labels in the training set were input into the neural network for classification training to obtain a model. The validation set was used to simultaneously evaluate the performance of the model during the training process, which helped to avoid overfitting of the model. The test set provided a completely independent portion of data to evaluate the performance of the model after training, reflect the actual effect of the model in practical applications, and determine the best model for application. The concentration of Fusarium in the sample was determined by the classification results of the fluorescent images output by the model;

[0099] S4. For the potato samples detected to have potato dry rot in step S3, potato tuber images are collected and sent to a U-Net-based segmentation model to determine the severity level of the potato disease.

[0100] Furthermore, the U-Net-based segmentation model used in step S4 classifies all pixels in the image into three categories: lesions, healthy tubers, and background. The total number of categories is defined as k, k = 3. The confusion matrix p is introduced to represent the classification results of all pixels in the tuber image, and p is used to represent the classification results of all pixels in the tuber image. ij Represents pixels whose true category is i and predicted by the model as j. The following three evaluation indicators are used to evaluate the performance of the segmentation model:

[0101] The average intersection over union (mIoU) is used to measure the overlap between the pixels in the real area and the pixels in the predicted area. The calculation method is shown in the following formula:

[0102]

[0103] The set similarity measurement function Dice is used: an indicator to measure the similarity between the segmented pixel samples predicted by the model and the real pixel samples. The calculation method is shown in the following formula:

[0104]

[0105] The global accuracy GC is used to calculate the ratio of all correctly predicted pixels to all pixels in a tuber image. The calculation method is as follows:

[0106]

[0107] Furthermore, when the U-Net-based segmentation model is used for grading in step S4, the tuber image is input into the trained U-Net-based segmentation model, the area ratio of the lesions is calculated according to the segmentation result, and the potato dry rot lesion grading standard expressed in symbolic form is integrated into the model. The model finally performs severity grading based on the lesion area ratio and outputs it.

[0108] Furthermore, the image data was obtained by taking photos with a mobile phone. The tuber sections were photographed and a potato dry rot image dataset annotated by professional researchers was established, which was divided into training set, validation set and test set according to the ratio of 7:1.5:1.5. The dry rot classification standard was determined by professionals, and the standard proposed by the 2004 China Plant Disease Chemical Control Academic Seminar and the Chinese Society of Plant Pathology on the control of potato tuber dry rot during storage was used. The dry rot tuber images were divided into 5 levels: Level 1: The lesion area accounts for less than 5% of the entire tuber area; Level 2: The lesion area accounts for 6%-15% of the entire tuber area; Level 3: The lesion area accounts for 16%-30% of the entire tuber area; Level 4: The lesion area accounts for 31%-50% of the entire tuber area; Level 5: The lesion area accounts for more than 50% of the entire tuber area.

[0109] First, a lesion segmentation model is established based on UNet and trained using the training set to obtain and save the trained segmentation model. During the experiment, the experimental parameters shown in Table 1 are set to the same.

[0110] Table 1 Segmentation parameters

[0111]

[0112] Furthermore, the severity of dry rot of the same potato was graded. Neural symbolic learning includes a neural network part and a symbolic part, which has higher interpretability than the general "black box" deep learning model. Figure 5 As shown in the figure, firstly, a deep learning model suitable for potato dry rot image segmentation is established based on UNet to segment the tuber section image. The model calculates the area ratio of the tuber section occupied by the lesion according to the segmentation result. At this time, the grading standard represented by symbols is integrated into the model, and the severity level of the disease is determined by comparing it with the area ratio represented by the same symbols. This neural symbolic learning model has higher grading accuracy and interpretability, and can achieve fast and accurate lesion segmentation and severity grading of potato dry rot. Figure 6As shown, Encoder (left half): The image passes through two convolutional layers with kernel size = 3×3 (Batch Normalization, ReLU activation) five times. After each convolutional layer, the image passes through a maximum pooling layer with pool size = 2×2, which is equivalent to the length and width of the image becoming 1 / 16 of the original image. Decoder (right half): The feature map that has passed four maximum poolings will pass through two convolutional layers with kernel size = 3×3 (BatchNormalization, ReLU activation) four times again. After each convolutional layer, the image will be upsampled once (the upsampling operation can be implemented in a variety of ways, such as using upsampling convolution or Bilinear Interpolation). In the process of passing through the decoder, the feature map is concat spliced ​​with the feature map of the same format in the encoder to form a thicker feature.

[0113] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0114] Although the present application has been described above with reference to specific embodiments, various modifications may be made thereto and parts thereof may be replaced with equivalents without departing from the scope of the present application. In particular, as long as there is no structural conflict, the various features in the specific embodiments disclosed in the present application may be used in combination with each other in any manner, and the fact that these combinations are not exhaustively described in this specification is only for the sake of omitting space and saving resources. Therefore, the present application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A smartphone fluorescence data acquisition device, characterized in that: It comprises an oblique incident lighting module (1), an intelligent control module (2), a fluorescence detection module (3), and a data storage module (4); The oblique-incidence illumination module (1) comprises an excitation light source (1-1), a first lens (1-2), a first reflector (1-3), a second reflector (1-4), and a second lens (1-5); the intelligent control module (2) comprises an intelligent controller (2-1), a first stepper motor (2-2), and a second stepper motor (2-3); the fluorescence detection module (3) comprises a stage (3-1), a triplet lens (3-2), a filter (3-3), and a mobile phone camera (3-4); The LED excitation light emitted by the excitation light source (1-1) is irradiated to the first reflector (1-3) through the first lens (1-2), then reflected to the second reflector (1-4), and then incident on the sample on the stage (3-1) through the second lens (1-5), and then the fluorescence signal emitted by the sample is collected through the triplet lens (3-2), and the sample fluorescence signal light is obtained through the filter (3-3), and the mobile phone camera (3-4) takes a fluorescent image and transmits it to the data storage module (4); The intelligent controller (2-1) is respectively connected to a first stepper motor (2-2) and a second stepper motor (2-3); the first stepper motor (2-2) is connected to a stage (3-1); and the second stepper motor (2-3) is connected to a filter (3-3).

2. A smartphone fluorescence data acquisition device according to claim 1, characterized in that: The first reflector (1-3) and the second reflector (1-4) are neutral attenuation reflectors, the first lens (1-2) is an aspheric lens, and the second lens (1-5) is a broadband achromatic lens; the excitation light source (1-1) is an LED excitation light source with a wavelength of 532nm; the obtained alignment light is irradiated onto the sample through the second lens (1-5) in a 75° illumination mode.

3. A smartphone fluorescence data acquisition device according to claim 2, characterized in that: The pixel of the mobile phone camera (3-4) is 20Mp.

4. A smartphone fluorescence data acquisition device according to claim 3, characterized in that: The first stepper motor (2-2) is used to precisely control the stage (3-1) with an accuracy of micrometers; the second stepper motor (2-3) is used to adjust the filter (3-3) to obtain a clear image of the sample taken in a clear focal plane, and the two motors are synchronously controlled by external triggering, and the operation of the motors is controlled by Arduino software.

5. A method for detecting potato dry rot, implemented by relying on a smartphone fluorescence data acquisition device according to any one of claims 1 to 4, characterized in that: The steps include: S1. Amplify the target nucleic acid and protein of potato based on RPA type III-A CRIPSR multi-stage signal amplification detection system to prepare potato dry rot detection samples; S2. placing a potato dry rot test sample on a stage of a smartphone fluorescence data acquisition device, starting the smartphone fluorescence data acquisition device, and collecting a fluorescence image of the potato dry rot test sample; S3. Inputting the potato dry rot fluorescence image obtained in step S2 into the trained neural network model for feature recognition, outputting the feature as the detected Fusariales nucleic acid concentration, and obtaining the potato dry rot detection result; S4. For the potato samples detected to have potato dry rot in step S3, potato tuber images are collected and sent to a U-Net-based segmentation model to determine the severity level of the potato disease.

6. A method for detecting potato dry rot according to claim 5, characterized in that: The specific implementation method of step S1 includes extracting DNA from potatoes, amplifying the target DNA fragment by RPA, and re-transcription to obtain an RNA test sample, and then reacting with 250 μM ATP, 500 nM fluorescent reporter RNA, 500 nM TtCsm complex and 2500 nM TtCsm6 in a reaction buffer with a 30 μL reaction system, and incubating the reaction at 60°C to obtain a potato dry rot test sample, wherein the reaction buffer includes 20 mM Tris-HCl with a pH of 7.9, 200 mM monopotassium glutamate, 10 mM ammonium sulfate, 5 mM magnesium sulfate, and 1 mM TCEP.

7. A method for detecting potato dry rot according to claim 6, characterized in that: The trained neural network model used in step S3 is a Transformer neural network. The specific training process includes the following steps: S3.

1. Collect 200 sets of fluorescence images of each concentration gradient, remove the noise of the negative control group and each concentration group by limiting the contrast adaptive histogram equalization method, and then divide them into training set, validation set and test set according to the ratio of 3:1:1; for the training set, perform data enhancement on the original image data by random flipping, mirroring and Gaussian smoothing methods, and finally convert all images into grayscale images; S3.

2. The neural network model is constructed as a Transformer neural network, which is mainly composed of an input part, a multi-layer encoder, a multi-layer decoder, and an output part. The setting includes N layers of encoders and N layers of decoders, and N in Transformer is 6; Set each encoder layer to have two sub-layers, namely the Multi-Head Attention layer and the Position-wise Feed-Forward Network, with a residual connection and a layer normalization LayerNorm operation behind each sub-layer; Each decoder layer has three sub-layers, Masked Self-Attention layer, Encoder-Decoder attention layer, and position-by-position feedforward neural network, with residual connection and layer normalization LayerNorm operation after each sub-layer; The attention mechanism calculates the attention score by matching the query tensor Q with the key tensor K, converts the score into the weighted value tensor V, and obtains the final attention vector. The attention vector calculation formula is as follows: Among them, softmax is the attention weight, is the scaling factor, d k is the dimension size of Q or K; In the multi-head attention mechanism, h attention heads run in parallel, and each attention head has an attention query matrix, a key matrix, and a value matrix. And output, then concatenate the outputs of all attention heads, and pass through the weight matrix W o Complete the linear transformation and get the final output. The calculation formula is as follows: MultiHead(Q,K,V)=Concat(head1,…,head h )W o S3.

3. Input the training set obtained in step S3.1 into the Transformer neural network constructed in step S3.

2. The label of the data set is the initial concentration of Fusarium before amplification. Observe whether the network output results are referenceable. In this process, continuously optimize the parameters through back propagation to obtain the optimal operation parameters and predicted values. In the model improvement part, observe the evaluation effect of the test set on the training model. According to the evaluation results, continuously optimize the network depth and related parameter settings to obtain a trained neural network.

8. A method for detecting potato dry rot according to claim 7, characterized in that: The U-Net-based segmentation model used in step S4 classifies all pixels in the image into three categories: lesions, healthy tubers, and background. The total number of categories is defined as k, k = 3. The confusion matrix p is introduced to represent the classification results of all pixels in the tuber image, and p is used to represent the classification results of all pixels in the tuber image. ij Represents pixels whose true category is i and predicted by the model as j. The following three evaluation indicators are used to evaluate the performance of the segmentation model: The average intersection over union (mIoU) is used to measure the overlap between the pixels in the real area and the pixels in the predicted area. The calculation method is shown in the following formula: The set similarity measurement function Dice is used: an indicator to measure the similarity between the segmented pixel samples predicted by the model and the real pixel samples. The calculation method is shown in the following formula: The global accuracy GC is used to calculate the ratio of the number of correctly predicted pixels to the number of all pixels in a tuber image. The calculation method is as follows:

9. A method for detecting potato dry rot according to claim 8, characterized in that: When the U-Net-based segmentation model is used for grading in step S4, the tuber image is input into the trained U-Net-based segmentation model, the area ratio of the lesions is calculated according to the segmentation result, and the potato dry rot lesion grading standard expressed in symbolic form is integrated into the model. The model finally performs severity grading based on the lesion area ratio and outputs it.

Images