Small molecule probe based on fluorescence sensing and application

Abstract

The invention relates to a small molecular probe TPE-J based on fluorescence sensing and application thereof. According to the intelligent real-time monitoring system for the nitro explosive, a PPYOLO algorithm is utilized, and a fluorescent colorimetric image is captured under the assistance of an optical camera. The fluorescent probe TPE-J is designed and synthesized and is used for dose sensitivity and visual detection of nitro explosives. Electron transfer between PA and the probe causes specific response, original blue fluorescence is rapidly quenched to be non-luminous within 5 seconds, and the detection limit is as low as 1 mg / mL. Color changes can be integrated into optical camera capture and quantization, and generated image data is automatically processed by a deep learning algorithm platform. The sensing system is beneficial to efficient real-time monitoring and high-sensitivity PA detection in various scenes. A fluorescence sensing detection platform combined with deep learning provides a new visual angle for efficient and portable detection of explosives.

Description

Technical Field

[0001] The present invention relates to the technical field of fluorescent probes, in particular to a small molecule probe based on fluorescent sensing and its application. Background Art

[0002] Although fluorescence sensing technology has the advantages of fast response speed, easy operation, and visualization, and is considered to be more suitable for on-site mobile detection needs (more suitable for civil and police scenarios), the technology still relies on precision optical equipment and well-trained technicians, which limits their application in mobile detection scenarios. In addition, complex spectral information further increases the detection time, and it is impossible to quickly obtain accurate trace quantitative detection results. Compared with the complex spectral data monitoring process, especially before long-term real-time recording and analysis, converting spectral data results into image information output is conducive to accurate and rapid data visualization. However, how to reduce human resource consumption, improve detection speed, accuracy and portability while achieving rapid data visualization has limited the further development of nitro explosive fluorescence sensing technology.

[0003] Deep learning enables machines to learn from large amounts of data to replicate human intelligence and has proven to be one of the most effective tools for image data processing. The use of fluorescent sensing technology to detect explosives generates a large amount of fluorescent image data, which is inefficient to process with traditional RGB value reading methods and greatly reduces accuracy. In contrast, deep learning, especially convolutional neural networks (CNNs), excels in automatically extracting image features, can recognize complex patterns and structures in images, and has superior performance in terms of speed and accuracy of image processing. Summary of the invention

[0004] The purpose of this application is to provide a method to solve the problems existing in the prior art.

[0005] The structure of the small molecule probe provided in the present application is as follows:

[0006]

[0007] The synthesis method of the small molecule probe is as follows: degas a mixture of 3,4-dibromothiophene, (4-(1,2,2-triphenylvinyl)phenyl)boric acid, toluene, potassium carbonate, distilled water, tetrakis(triphenylphosphine)palladium and anhydrous ethanol, and stir and reflux under nitrogen at 90° C. for 24 hours to obtain the obtained product.

[0008] The reaction molar ratio of 3,4-dibromothiophene, (4-(1,2,2-triphenylvinyl)phenyl)boric acid and potassium carbonate is 1:3:8; the volume ratio of toluene, distilled water and anhydrous ethanol added is 6:4:3; the dosage ratio of toluene is 0.75 mL of toluene per mmol of potassium carbonate; the equivalent of tetrakis(triphenylphosphine)palladium added is 0.02 times the equivalent of 3,4-dibromothiophene.

[0009] Application of small molecule probes in detecting explosives containing picric acid.

[0010] A fluorescent sensor, on which the small molecule probe is fixed.

[0011] Furthermore, the fluorescence sensor is a paper-based fluorescence sensor or a hydrogel fluorescence sensor film.

[0012] A portable explosive detection platform comprises a laptop computer, a closed box, a UV light source, an optical camera and a sample bottle, wherein a fluorescent sensor or a small molecule probe is installed in the sample bottle.

[0013] A method for quantifying picric acid content based on fluorescence image spectrum, using a portable explosive detection platform or the fluorescence sensor or the small molecule probe to combine with picric acid to obtain an image photographed by an optical camera, extracting RGBV values ​​in the image, and establishing a linear relationship between RGB and HSV values ​​and picric acid concentration. The qualitative and concentration of picric acid can be detected based on the fluorescence image calculation through a linear relationship formula.

[0014] Furthermore, the image is processed using the PPYOLO model, which receives the input image and uses a deep convolutional neural network for feature extraction; a feature pyramid network and a path aggregation network are used to enhance the representation of multi-scale features; then, the network processes the feature map to predict the category probability and bounding box coordinates of each region; anchor boxes are used to match the target, and non-maximum suppression is applied to eliminate overlapping predictions; finally, a post-processing step is performed to filter and fine-tune the results to obtain the final object detection result.

[0015] A cloud-based intelligent visualization detection system is provided, wherein the PPYOLO model is deployed on a cloud platform to obtain a cloud-based intelligent visualization detection system.

[0016] The beneficial effects of the present invention are as follows: the intelligent real-time monitoring system for nitro explosives of the present invention utilizes the PPYOLO algorithm to capture fluorescent colorimetric images with the assistance of an optical camera. A fluorescent probe TPE-J was designed and synthesized for dose-sensitive and visual detection of nitro explosives (PA). The electron transfer between PA and the probe induces a specific response, rapidly quenching the original blue fluorescence to non-luminescence within 5 seconds, with a detection limit as low as 1 mg / mL. The color change can be integrated into an optical camera to capture and quantify, and the resulting image data is automatically processed by a deep learning algorithm platform. The sensing system facilitates efficient real-time monitoring and high-sensitivity detection of PA in a variety of scenarios. The fluorescence sensing detection platform combined with deep learning provides a new perspective for efficient portable detection of explosives. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is the hydrogen spectrum detection map of the small molecule probe TPE-J.

[0018] Figure 2 This is the carbon spectrum detection spectrum of the small molecule probe TPE-J.

[0019] Figure 3 This is the synthetic route of the small molecule probe TPE-J.

[0020] Figure 4 This is the Stern–Volmer plot of I0 / I-1 and picric acid concentration of the small molecule probe TPE-J point.

[0021] Figure 5 This is the relationship between the fluorescence intensity and wavelength of the small molecule probe TPE-J and picric acid.

[0022] Figure 6 This is the relationship between the quenching intensity and time of the small molecule probe TPE-J and picric acid.

[0023] Figure 7 This is a darkroom imaging photograph of the fluorescence quenching reaction between the small molecule probe TPE-J and picric acid under ultraviolet light.

[0024] Figure 8 This is the DFT theoretical calculation structure diagram of the small molecule probe TPE-J and picric acid.

[0025] Fig. 9 This is the reaction mechanism of the small molecule probe TPE-J and picric acid.

[0026] Fig.10 This is a flow chart of the method for processing fluorescence images based on the PPYOLO model.

[0027] Fig.11 The RGBV value map obtained by using image processing method under different picric acid concentrations.

[0028] Fig.12 This is a comparison chart between the analyte concentration values ​​detected by the image processing method and the actual analyte concentration values ​​at different picric acid concentrations.

[0029] Fig.13 The linear relationship diagram of RGBV values ​​obtained by image processing method under different picric acid concentrations. a, b, c, d represent the linear relationship between picric acid concentration and photon fluorescence channel R, G, B, V value information, respectively. e represents the functional relationship between picric acid concentration and photon fluorescence channel value information ΔE. f represents the comparison between the analyte concentration value detected by RGBVE and the actual analyte concentration value. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] In the following examples, unless otherwise specified, the "water" used is deionized water.

[0032] In the following tests of the present invention, nuclear magnetic resonance hydrogen spectrum tests were performed on an AVANCE NEO 400MHz nuclear magnetic resonance spectrometer produced by Bruker, Germany. The solvent used was deuterated chloroform (CDCl3) and the calibration was carried out with tetramethylsilane (TMS).

[0033] Example 1. Preparation of small molecule probe TPE-J and study of its fluorescence quenching response to picric acid

[0034] 1. The synthesis method of small molecule probe TPE-J includes:

[0035] A mixture of 3,4-dibromothiophene (483.86 mg, 2 mmol), (4-(1,2,2-triphenylvinyl)phenyl)boric acid (2.75 g, 6 mmol), toluene (12 mL), potassium carbonate (2.21 g, 16 mmol), distilled water (8 mL), tetrakis(triphenylphosphine)palladium (92.40 mg, 0.04 mmol) and anhydrous ethanol (6 mL) was degassed and stirred under reflux at 90 °C for 24 h under nitrogen to prepare a small molecule probe with fluorescence response. Figure 1 ) and carbon spectrum ( Figure 2 ) The structure of the small molecule probe (TPE-J) is as follows:

[0036]

[0037] The synthetic route is Figure 3 As shown. The reaction molar ratio of 3,4-dibromothiophene, (4-(1,2,2-triphenylvinyl)phenyl)boric acid and potassium carbonate is 1:3:8; the solvent used is toluene, distilled water and anhydrous ethanol, the volume ratio is 6:4:3, the amount of toluene is 0.75mL per millimole of potassium carbonate; the equivalent of tetrakis(triphenylphosphine)palladium added is 0.02 times the equivalent of 3,4-dibromothiophene.

[0038] 2. Study on the fluorescence quenching response of small molecule probe TPE-J to picric acid

[0039] The synthesized probe TPE-J was diluted to 1 μM with THF tetrahydrofuran and subjected to fluorescence spectroscopy experiments. Time-dependent UV-visible spectroscopy of TPE-J and picric acid was studied in a tetrahydrofuran aqueous solution (tetrahydrofuran: water (volume ratio) = 1:9). The UV-visible absorption spectrum was measured on a Hitachi UV-3900 UV-visible spectrophotometer.

[0040] The results are as follows: The linear relationship between fluorescence quenching intensity and picric acid concentration is as follows Figure 4 As shown, the concentration is 1 μM, I represents the peak intensity, and I0 represents the peak intensity without PA. The relationship between fluorescence intensity and wavelength is shown in Figure 5 As shown, the emission wavelength of TPE-J is determined. Figure 6 As shown in the figure, the reaction of TPE-J with picric acid at a final concentration of 40 μg / mL basically reached equilibrium within 30 seconds through kinetic measurement. Fluorescence quenching visualization was achieved by darkroom imaging. The fluorescence image under 365 nm ultraviolet light in the darkroom is shown in the figure. Figure 7 shown.

[0041] Gaussian 16 program package was used to perform DFT (density functional theory) calculations on probe TPE-J and picric acid PA. Without CPCM solvation model, the B3LYP / 6-31G (d, p) energy level calculation method was used, and the calculation results are as follows: Figure 8As shown, the molecular orbital (MO) diagram and MO energy level were calculated at the same theoretical level. In the geometry optimization of the two molecules, the HOMO-LUMO gaps of TPE-J and picric acid were 3.93 eV and 5.45 eV, respectively. In addition, PA exhibits a wider band gap and lower LUMO than TPE-J. In general, nitro compounds with relatively low LUMO energy levels can accept excited-state electrons from AIE molecules, thereby quenching the fluorescence of AIE molecules. The LUMO energy gap between the AIE sensor and the nitro compound is regarded as the "driving force". The natural bond orbital method was used to study the electrostatic potential surface (EPS) of the probe to evaluate its preferential reaction site for PA. It was observed that the extrema of the probe were located in the thiophene ring and the adjacent benzene ring part of the molecule (red area), indicating that this area has a strong attraction to PA. Many polynitro-substituted aromatic explosives, which are electron-deficient in nature, can bind to electron-rich species through donor-acceptor interactions.

[0042] like Fig. 9 The reaction mechanism of the probe TPE-J and picric acid is as follows: the electron-rich group of the probe (TPE-J) and the electron-deficient picric acid undergo a photoinduced electron transfer reaction to achieve fluorescence response. The probe reacts with picric acid quickly, and the release of the picric acid solution makes the chemical reaction more complete. During the photoinduced electron transfer (PET) process, the excited fluorophore donor provides an electron acceptor to the LUMO through its LUMO. The electron-rich TPE-J interacts with the electron-deficient nitroaromatic molecule through electrostatic interaction, which is beneficial for high-sensitivity detection of explosives.

[0043] Example 2: Application of small molecule probe TPE-J in fluorescence sensor

[0044] The small molecule probe TPE-J is fixed on a carrier to prepare a fluorescent sensor that responds to picric acid explosive substances, including a paper-based fluorescent sensor and a hydrogel fluorescent sensor film.

[0045] The paper-based fluorescent sensor is prepared by loading the small molecule probe TPE-J with fluorescent response on a paper-based carrier through solution deposition. The hydrogel fluorescent sensor film is prepared by compounding the small molecule probe TPE-J with fluorescent response with agarose to obtain a hydrogel with a fluorescent probe, and the hydrogel three-dimensional porous structure is used to adsorb and detect picric acid solid particles, and has good mechanical properties.

[0046] The structural formula of agarose is as follows:

[0047]

[0048] The method for preparing a hydrogel film with a fluorescent small molecule probe comprises: mixing a small molecule probe TPE-J with a fluorescent response with an agarose hydrogel at a mass ratio of 1:200, reacting in water for 5 minutes, and preparing a hydrogel with a fluorescent probe.

[0049] Example 3: Portable nitro explosive PA detection platform using small molecule probe TPE-J

[0050] The portable detection platform includes a laptop computer, a closed box, a UV light source, an optical camera and a sample bottle. The small molecule probe TPE-J or a fluorescent sensor composed of the small molecule probe TPE-J is fixed in the sample bottle. The reaction is carried out in a closed box, which can eliminate the influence of external light sources on fluorescent sensing. The picric acid-containing substance is put into the sample bottle and photographed under a UV light source. The fluorescent image is obtained by optical camera photography. By analyzing the image color information and parsing the spectral information, accurate identification and quantitative detection of picric acid explosives are obtained.

[0051] The color information includes red: R, green: G, blue: B, hue: H, saturation: S, and brightness: V. The captured image is segmented and extracted according to these six image signal channels, and the relationship between RGB and HSV values ​​and the PA concentration in the sample is established. Thus, the PA concentration in the sample can be obtained through the RGB and HSV values ​​of the image. This part uses the PP-YOLO model.

[0052] like Fig.10 Image processing process shown: The PPYOLO model receives an input image and uses a deep convolutional neural network for feature extraction. It uses a feature pyramid network (FPN) and a path aggregation network (PANet) to enhance the representation of multi-scale features. Subsequently, the network processes the feature map to predict the class probability and bounding box coordinates for each region. The target is matched using anchor boxes, and non-maximum suppression (NMS) is applied to eliminate overlapping predictions. Finally, post-processing steps are performed to filter and fine-tune the results to obtain the final object detection results.

[0053] Specifically, it includes the following parts:

[0054] 1. Input processing: Image preprocessing includes adjusting the image size to a fixed size and normalizing it to meet the needs of the network; during the training phase, data augmentation techniques such as random cropping, rotation, and flipping are applied to improve the generalization ability of the model.

[0055] 2. Backbone network: PPYOLO uses ResNet50 or similar backbone networks as feature extractors. These networks usually include multiple convolutional layers and pooling layers to capture optical features in images.

[0056] Residual connections: Skip connections such as those introduced in ResNet can help solve the vanishing gradient problem in deep networks and promote information flow. Backbone uses ResNet50-vd, which has powerful feature extraction capabilities, to extract feature maps of different scales. In addition, the introduction of deformable convolutional networks (DCN) in ResNet50-vd can better handle the deformation and posture changes of the target, help improve the accuracy of target recognition, and enhance the model's adaptability to the target. This improvement enables PPYOLO to achieve high accuracy while maintaining high efficiency, making it suitable for target detection tasks in a variety of scenarios.

[0057] 3. Feature Pyramid Network (FPN): FPN constructs multi-scale feature representation by upsampling feature maps at different levels and fusing information from adjacent levels. It can detect different targets at the same time. FPN can effectively extract multi-scale features to meet the requirements of target detection across different scales. After obtaining multi-level feature maps from the backbone, three resolution feature levels are formed. After images of different scales are input from the 1*1 convolution layer, a top-down path is first constructed. Upsample Block is used in the backbone network to upsample high-level (shallower but richer in semantics) feature maps to match the size of low-level (deeper but richer in space) feature maps. Horizontal connections are introduced, and the upsampled features are fused with the corresponding low-level feature maps using 1*1 convolution layers to match the channel size. The generated feature maps after fusion are output for the next step of detection and prediction. The fusion of high-level semantic features and low-level spatial features finally obtains multi-scale feature maps, which are output through DepthwiseSeparable Convolution.

[0058] 4. Path Aggregation Network (PANet): PANet further enhances the effect of FPN by adding additional bottom-up paths to strengthen the information transfer from low layers to high layers.

[0059] 5. Head prediction: Add classification head and regression head on the feature maps of different scales generated by FPN / PANet.

[0060] Classification head: responsible for predicting the probability of an object class present at each position.

[0061] Regression head: gives the position offset relative to the anchor box and the aspect ratio change to accurately locate the target. PPYOLO's detection head consists of a 3×3 convolutional layer and a 1×1 convolutional layer to improve feature extraction and channel number adjustment for final prediction. The output channels of each final prediction are 3(K+5), where K is the number of categories. Each position on each final prediction map is associated with three different anchors in the next section. For each anchor, the first is the prediction of the category probability of the detection head for each position, and the probability prediction of the Kth class is output by the first K channels of the convolutional layer. The four channels in the second part are responsible for the position prediction of the bounding box, which is completed by the output of the four channels in the middle of the convolutional layer. The last channel is the objective score prediction, which usually uses the model to predict quantifiable results based on objective standards. The accuracy of these predictions can be verified by comparing with the true values.

[0062] 6. Loss function: The PPYOLO model uses different types of loss functions to train its classification and positioning tasks. In order to classify the target and train the model to improve the accuracy of category prediction, Cross Entropy Loss is used to measure the difference between the category probability distribution predicted by the model and the true category; L1 Loss is sensitive to the absolute value of the distance and is used to measure the distance between the predicted bounding box and the true bounding box to accurately locate the target; Objectness Loss is used to monitor whether the model recognizes the existence of the target object. The calculation of Objectness Loss is usually implemented through binary cross entropy loss, which quantifies the difference between the model's prediction of the existence of the object and the actual scene. This multi-task learning method enables the model to complete the target detection task quickly and accurately.

[0063] 7. Post-processing: threshold filtering, bounding box adjustment, class probability assignment, score re-ranking, output formatting (img / csv) and making the results visible.

[0064] 8. Output results: Final output: The PPYOLO model optimizes the final test results of the model through Postprocess and selects the prediction results with higher credibility. The algorithm formats the output results into the required output format (csv or jpg) and visualizes the detection results (labeling the image with bounding boxes and category labels). Through the post-processing step, PPYOLO ensures that the model can provide high-quality object detection results suitable for various practical application scenarios.

[0065] Through the above process, PPYOLO can achieve fast target detection while ensuring high accuracy. This design is particularly suitable for application scenarios that require real-time performance, such as video surveillance, autonomous driving and other fields. In this embodiment, the color change of the probe solution in the sample bottle is monitored at a rate of 0.1s per frame. The operator manually selects the target identification area in the image, and the color extraction function outputs the RGB / HSV value. In addition, the image data can be automatically recorded and stored according to the timeline to construct a complete data set as the input of the subsequent deep learning platform.

[0066] PA samples with different concentrations (0, 10, 20, 30 μg / mL) were input into the PA detection platform for accuracy verification. Fig.11 and Fig.12 As shown in the figure, the platform can extract the R, G, B, V and E values ​​at each concentration, and compare the analyte concentration value detected by the platform with the true analyte concentration value, with an accuracy of more than 95%. The fluorescence quenching concentration curve of PA was further analyzed to establish the relationship between the image information and the target concentration. In addition to deriving the RGB signal value from the fluorescence image output, the HSV signal value is generated by adding a signal output channel to improve the image processing accuracy.

[0067] After the above steps, the same color data can be retrieved from the same image on multiple image processing platforms such as mobile phone applications, computer software (Photoshop CC 2018), and cameras, confirming that the extraction of RGBV values ​​is not affected by the differences between different image processing tools. In scenarios where there is no dedicated camera, smartphone cameras can also perform recognition, thus meeting a wider range of application needs.

[0068] The relationship between RGB and HSV values ​​and PA concentration was determined by calibrating the standard curve. The linear relationship between RGB and HSV signal values ​​and target concentration was analyzed experimentally, and the results are shown in Table 1 and Fig.13 As shown in Figure 2, the R, G, B, and V values ​​all showed a good linear correlation with the PA concentration (R 2 >99). This indicates that the detection platform can accurately determine the RGBV value through image analysis, and thus the PA concentration. It was found that the correlation between color information and target concentration showed two different stages. Initially, when the PA concentration was in the range of 0 to 10 μmg / mL, the degree of fluorescence quenching was obvious, resulting in a significant change in the RGBV value of the image. Subsequently, as the PA concentration continued to increase outside this range, the degree of fluorescence quenching changed slowly, and the relationship between the RGBV value and the PA concentration showed another linear relationship.

[0069] In addition, PA concentration can be detected by visual color perception. Changes in visual color perception can be quantified using the following formula:

[0070]

[0071] Among them, Rn, Bn, Gn, Vn represent the R, B, G, V values ​​after adding quenching agent, and R0, B0, G0, V0 represent the R, B, G, V values ​​without adding quenching agent. Fig.13 As shown in e.

[0072] Table 1 Evaluation results of different concentrations of PA

[0073]

[0074] Example 4: Cloud-based intelligent visual detection system

[0075] The method of color analysis of fluorescent images in the PA detection platform in Example 3 is deployed on the cloud platform, so that multiple detection platforms and multiple detection locations can be covered through the open interface of the cloud platform, so that multiple detection terminals are standardized and real-time data transmission and data sharing can be achieved, which helps to establish a huge database and provide valuable resources for future model training and updating. In addition, multiple mobile terminals can be connected to the cloud platform by establishing a cloud website, which increases the ways to detect explosives. The PPYOLO model is trained and automatically learned and updated based on the data set collected in the cloud, improving the efficiency and accuracy of target detection.

[0076] In this embodiment, the open source platform selected PaddlePaddle, and the PPYOLO target detection model based on the PaddlePaddle deep learning platform represents an enhanced version of the YOLO (You Only Look Once) algorithm. The YOLO algorithm is widely praised for its speed and performance, and is particularly suitable for real-time processing scenarios, effectively solving the actual needs of PA detection. Based on YOLO, PPYOLO has been optimized and upgraded to improve detection accuracy and speed. In this embodiment, the intelligent real-time monitoring system of nitro explosives uses the PPYOLO algorithm to capture fluorescent colorimetric images with the assistance of an optical camera. A fluorescent probe (TPE-J) composed of a TPE fragment and a thiophene [3,4-b] group was designed and synthesized for dose-sensitive and visual detection of nitro explosives (PA). The electron transfer between PA and the probe causes a specific response, rapidly quenching the original blue fluorescence to non-luminescence within 5 seconds, with a detection limit as low as 1 mg / mL. Using density functional theory (DFT), the competitive reaction mechanism between the detector and the fluorescent sensor was explored. The color change can be integrated into the optical camera to capture and quantify, and the resulting image data is automatically processed by the deep learning algorithm platform. The sensing system facilitates efficient real-time monitoring and high-sensitivity detection of PA in various scenarios. The fluorescent portable sensing platform combined with deep learning provides a new perspective for efficient portable detection of explosives.

[0077] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be considered as exemplary and non-restrictive in all respects.

Claims

1. A small molecule probe based on fluorescence sensing, characterized in that: The structure of the small molecule probe is shown below:

2. A method for synthesizing a small molecule probe as claimed in claim 1, characterized in that: The synthesis method of the small molecule probe is as follows: degas a mixture of 3,4-dibromothiophene, (4-(1,2,2-triphenylvinyl)phenyl)boric acid, toluene, potassium carbonate, distilled water, tetrakis(triphenylphosphine)palladium and anhydrous ethanol, and stir and reflux under nitrogen at 90° C. for 24 hours to obtain the obtained product.

3. The synthesis method according to claim 2, characterized in that The reaction molar ratio of 3,4-dibromothiophene, (4-(1,2,2-triphenylvinyl)phenyl)boric acid and potassium carbonate is 1:3:8; the volume ratio of toluene, distilled water and anhydrous ethanol added is 6:4:3; the dosage ratio of toluene is 0.75 mL of toluene per mmol of potassium carbonate; the equivalent of tetrakis(triphenylphosphine)palladium added is 0.02 times the equivalent of 3,4-dibromothiophene.

4. Use of the small molecule probe as claimed in claim 1 in detecting explosives containing picric acid.

5. A fluorescence sensor, characterized in that: The small molecule probe as claimed in claim 1 is fixed on the fluorescent sensor.

6. The fluorescence sensor according to claim 5, characterized in that: The fluorescent sensor is a paper-based fluorescent sensor or a hydrogel fluorescent sensor film.

7. A portable explosives detection platform, characterized in that: The device comprises a laptop computer, a closed box, a UV light source, an optical camera and a sample bottle, wherein the sample bottle is filled with the fluorescent sensor as claimed in claim 5 or the small molecule probe as claimed in claim 1.

8. A method for quantifying picric acid content based on fluorescence image spectrum, characterized in that: The portable explosive detection platform as claimed in claim 7, the fluorescent sensor as claimed in claim 5, or the small molecule probe as claimed in claim 1 is used to combine with picric acid and then photographed by an optical camera to obtain an image, extract the RGBV value in the image, and establish a linear relationship between the RGB and HSV values ​​and the picric acid concentration. The qualitative and concentration of picric acid can be detected based on the fluorescence image through the linear relationship formula.

9. The method according to claim 8, characterized in that The image processing uses the PPYOLO model, which receives the input image and uses a deep convolutional neural network for feature extraction; Use feature pyramid network and path aggregation network to enhance the representation ability of multi-scale features; The network then processes the feature maps to predict class probabilities and bounding box coordinates for each region; matches objects using anchor boxes and applies non-maximum suppression to eliminate overlapping predictions; finally, a post-processing step is performed to filter and fine-tune the results to produce the final object detection results.

10. A cloud-based intelligent visualization detection system, characterized in that: The PPYOLO model as claimed in claim 9 is deployed on a cloud platform to obtain a cloud-based intelligent visualization detection system.

Images