Tomato early blight intelligent detection method and system
By integrating hyperspectral imaging, leaf underside microscopy, and polarization imaging technologies, the specific pathological characteristics of early blight in tomatoes are directly captured and quantitatively analyzed, achieving early and highly accurate diagnosis of early blight in tomatoes and solving the problems of low efficiency and misdiagnosis in existing technologies.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI PUSHA INVESTMENT DEVELOPMENT CO LTD
- Filing Date
- 2026-01-06
- Publication Date
- 2026-04-17
AI Technical Summary
Existing technologies are inefficient in detecting early blight in tomatoes, are not sensitive to early symptoms, and are greatly affected by ambient light. They cannot accurately identify features such as the black mold layer on the underside of leaves and early water-soaked chlorotic halo, leading to misdiagnosis and confusion.
By simultaneously acquiring hyperspectral images of the front of tomato leaves, microscopic images of the back of leaves, and polarization optical information of the front of leaves, and by analyzing near-infrared reflectance, morphological analysis, and polarization reflection characteristics, combined with a set of logical judgment rules based on plant disease knowledge, accurate diagnosis of early blight in tomatoes can be achieved.
It enables early and highly accurate diagnosis of early blight in tomatoes, advancing the detection window by 3-5 days, solving the problem of disease confusion, and ensuring the stability and reliability of the test results.
Smart Images
Figure CN121877754A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of smart agriculture technology, and in particular to a smart detection method and system for early blight of tomatoes. Background Technology
[0002] Early blight of tomatoes is caused by Alternaria alternata, and its typical symptoms are concentric ring-shaped lesions on the upper surface of leaves and a black mold layer on the underside. Current field detection mainly relies on manual observation, which is inefficient and insensitive to early symptoms. Some current so-called "intelligent detection" solutions are essentially simple applications of general image recognition algorithms in agriculture, and have fundamental flaws.
[0003] 1. Only analyzing the visible light images of the front of the leaf, completely ignoring the diagnostic feature of the black mold layer on the back of the leaf, leads to confusion with leaf blight, pesticide damage, etc.
[0004] 2. It can only identify obvious lesions that have expanded, but cannot capture early signs that are difficult to distinguish with the naked eye, such as water-soaked discoloration halos at the onset of the disease;
[0005] 3. It is greatly affected by ambient light; cloudy days and reflective light will cause feature extraction to fail.
[0006] Therefore, there is an urgent need for a specialized detection technology that can accurately identify the pathological characteristics of this disease from its early stages. Summary of the Invention
[0007] The purpose of this invention is to provide an intelligent detection method and system for optically capturing and judging the specific symptoms of early blight in tomatoes at each stage, so as to achieve early and highly accurate diagnosis and overcome the shortcomings of general image recognition methods.
[0008] To achieve the above objectives, the present invention adopts the following technical solution:
[0009] A smart detection method for early blight in tomatoes includes the following steps:
[0010] Simultaneously acquire hyperspectral image information of the front of the tomato leaf, microscopic image information of the back of the leaf, and polarization optical information of the front of the leaf for the same tomato leaf region;
[0011] From the aforementioned frontal hyperspectral image information, analyze the reflectance characteristics of specific near-infrared bands to extract the early halo index characterizing water-soaked lesions in cells;
[0012] Based on morphological analysis, the characteristics of black mold were identified from the microscopic images of the leaf back, and the morphological density of the mold layer on the back was calculated.
[0013] From the polarization optical information on the front side of the blade, the polarization reflection characteristics of the concentric ring pattern structure are analyzed to obtain the polarization contrast of the concentric ring pattern.
[0014] The early halo index, the morphological density of the mold layer on the back, and the polarization contrast of the concentric lines are input into a set of logical judgment rules based on plant disease knowledge, and the diagnostic conclusions for early blight of tomato and its stages are directly output.
[0015] Preferably, the extraction of the early halo index specifically involves:
[0016] Calculate the reflectance ratio of the blade region in the 780nm and 850nm bands;
[0017] Identify the closed, low-value ring-shaped regions appearing in the ratio graph;
[0018] The index value is calculated based on the extent to which the area and ratio of the annular region are below the threshold of the healthy region.
[0019] Preferably, the calculation of the morphological density of the mold layer on the back includes:
[0020] Threshold segmentation is performed on the microscopic images of the underside of leaves to extract dark connected components within a specific area range;
[0021] Analyze the circularity and edge texture roughness of each connected region to screen out regions that conform to the morphological characteristics of the early blight fungus mold layer;
[0022] After statistical screening, the number of regions per unit area and the area ratio are used to generate density parameters.
[0023] Preferably, the concentric fringe polarization contrast is obtained; specifically:
[0024] Calculate the degree of polarization image using frontal images obtained from different polarization directions;
[0025] On the polarization image, directional filtering or Radon transform is used to enhance the radial stripe features, and the strength of the directional consistency of these stripes is measured as a contrast parameter.
[0026] Preferably, the logical decision rule set includes at least the following rules:
[0027] If the density of the mold layer on the back exceeds the first diagnostic threshold, it is directly determined to be the middle to late stage of early blight in tomatoes;
[0028] If the morphological density of the mold layer on the back does not exceed the first diagnostic threshold, but the early halo index and the polarization contrast of the concentric lines both exceed their respective early warning thresholds, then it is determined to be early stage or suspected early blight of tomato.
[0029] A smart detection system for early blight of tomatoes that implements the method, comprising:
[0030] A probe-type acquisition head is used to simultaneously acquire image information of the front and back sides of a blade while it is being held.
[0031] The upper inner side of the acquisition head is integrated with a hyperspectral-polarization imaging module for acquiring frontal hyperspectral image information and providing polarization illumination.
[0032] The lower inner side of the acquisition head is integrated with a microscopic camera module for acquiring microscopic image information of the leaf back;
[0033] A processing and display unit connected to the acquisition head includes a built-in processor for running feature extraction algorithms and logical judgment rule sets, as well as a human-computer interaction interface for displaying diagnostic results.
[0034] Preferably, the hyperspectral-polarization imaging module includes: a hyperspectral imager covering the visible to near-infrared band, and a polarization illumination source consisting of at least three LED rings with different linear polarization directions, wherein the light source provides active illumination with a controllable polarization angle for the hyperspectral imager.
[0035] Preferably, the microscope imaging module includes: a fixed-focus microscope lens with a working distance of 1-3 cm, a high-resolution image sensor, and a set of LED supplementary lights to provide lateral illumination for the underside of the leaf.
[0036] Preferably, the inner edge of the clamping head of the probe is provided with a light-shielding sealing strip to isolate external ambient light interference when clamping the blade.
[0037] Preferably, the processing display unit is a handheld device, the logic decision rule set is stored in its memory in firmware form, and the processor is an FPGA or an embedded ARM chip.
[0038] The beneficial effects of this invention are:
[0039] Compared with existing technologies, this invention integrates hyperspectral imaging, leaf underside microscopy, and polarization imaging technologies to directly capture and quantify the specific pathological characteristics of early blight in tomatoes, achieving the following breakthroughs: First, near-infrared spectroscopy analysis enables ultra-early warning of water-soaked lesions during the incubation period, advancing the detection window by 3-5 days. Second, leaf underside microscopy directly captures the diagnostic feature of the pathogen's mold layer, fundamentally solving the problem of confusion with other similar diseases and significantly improving diagnostic specificity. Third, active polarization illumination technology effectively eliminates ambient light interference, ensuring the stability and reliability of the detection results. The entire system is integrated into a portable handheld device, enabling real-time diagnosis in the field and providing reliable technical support for precision prevention and control.
[0040] Other features and advantages of this application will be set forth in the following description and will be apparent in part from the description or may be learned by practicing embodiments of this application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings. Attached Figure Description
[0041] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0042] Figure 1 This is a flowchart of the intelligent detection method for early blight of tomatoes according to the present invention;
[0043] Figure 2 This is a block diagram of the intelligent detection system for early blight of tomatoes according to the present invention;
[0044] Figure 3 , Figure 4 This is a schematic diagram of the handheld detector used in the intelligent detection system for early blight of tomatoes of the present invention.
[0045] In the figure: handheld detector 1, upper panel 101, hyperspectral imaging module 1011, polarized light illumination module 1012, lower panel 102, supplementary light 1021, microscope camera 1022, sealing strip 103. Detailed Implementation
[0046] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.
[0047] The following description uses an example of a handheld detector.
[0048] Early characteristics (invasion stage - incubation stage): In the early stage of pathogen infection, water-soaked changes appear in the intercellular spaces of leaves. The reflectance in a specific near-infrared band (750-900nm) will show a characteristic decrease, forming an "invisible halo". This invention uses hyperspectral imaging to capture this characteristic.
[0049] Diagnostic features (symptom onset period): The black mold layer (conidiophores and spores) produced at the stomata on the underside of the leaf is the only diagnostic feature that distinguishes it from other leaf spot diseases. This invention uses back microscopic imaging to directly observe this morphology.
[0050] Typical characteristics (disease development stage): concentric ring patterns appear on the front, and the texture is different from the reflection characteristics of polarized light by the waxy layer on the surface of healthy leaves. This invention uses polarized imaging to enhance this texture and eliminate general discoloration interference.
[0051] refer to Figure 2 This is a block diagram of the intelligent detection system for early blight of tomatoes according to the present invention; Figure 3 , Figure 4 This is a schematic diagram of the handheld detector used in the intelligent detection system for early blight of tomatoes of the present invention; in this embodiment, the intelligent detection system for early blight of tomatoes is a handheld detector 1, comprising:
[0052] Clip-type acquisition head: It resembles a large clip and includes an upper panel 101 and a lower panel 102.
[0053] The inner side of the upper panel 101 is equipped with a hyperspectral imaging module 1011 (spectral range 500-1000nm) and a polarized light illumination module 1012 (which can emit LED light with different polarization angles).
[0054] Inside the lower panel 102: A high-resolution microscope camera 1022 (equipped with a megapixel CMOS, working distance of 2cm, and resolution up to 10μm / pixel) and its matching side fill light 1021 are arranged.
[0055] The inner edge of the clip has a sealing strip 103, which can be closed to block external light during collection.
[0056] Main controller: Embedded processing board (such as FPGA or high-performance ARM chip), controller is equipped with touch screen and battery.
[0057] refer to Figure 1 The flowchart of the intelligent detection method for early blight of tomatoes according to the present invention includes the following steps:
[0058] S1; Simultaneously acquire the front hyperspectral image information, back microscopic image information, and front polarization optical information of the same tomato leaf area;
[0059] S2; Analyze the reflectance characteristics of specific near-infrared bands from the aforementioned frontal hyperspectral image information, and extract the early halo index that characterizes water-soaked lesions in cells;
[0060] S3; Based on morphological analysis, identify the characteristics of black mold from the microscopic image information of the leaf back and calculate the morphological density of the mold layer on the back.
[0061] S4; From the polarization optical information on the front side of the blade, analyze the polarization reflection characteristics of the concentric ring pattern structure to obtain the polarization contrast of the concentric ring pattern.
[0062] S5; Input the early halo index, the morphological density of the mold layer on the back, and the polarization contrast of the concentric lines into the logical judgment rule set based on plant disease knowledge, and directly output the diagnostic conclusion for early blight of tomato and its stages.
[0063] The specific testing methods are as follows:
[0064] In-situ clamp sampling involves placing the tomato leaf to be tested between the open sampling heads, closing the probe clamps so that the leaf faces the hyperspectral and polarization imaging system on the upper panel, and the back of the leaf is pressed against the microscope lens protective window on the lower panel.
[0065] Synchronous acquisition of multimodal images, controlled by the processing board:
[0066] a) Turn on the polarized light illumination module and the hyperspectral imaging module continuously acquires hyperspectral image cube data of the front of the blade in three polarization directions: 0°, 45°, and 90°.
[0067] b) Simultaneously turn on the side supplement light on the back of the leaf, and the microscope camera captures a high-definition digital image of the back of the leaf;
[0068] Extraction of disease-specific characteristic parameters, and processing of the board to solidify the algorithm:
[0069] Parameter A (Early Halo Index): From the frontal hyperspectral data, the reflectance images of the 780nm and 850nm bands are extracted, the ratio of the two is calculated, and the edge detection algorithm is used to identify whether there are closed ring regions with a ratio lower than the healthy area threshold. If so, the area and average ratio difference of the ring region are calculated, and the "Early Halo Index" is obtained by combining the results.
[0070] Parameter B (morphological density of mold layer on the back): Adaptive binarization and morphological opening are performed on the microscopic image of the leaf back to extract all dark connected regions with an area between 100 and 5000 square micrometers. The roundness and edge roughness (morphological differences between true mold clusters and dust particles) of these connected regions are analyzed to screen out suspected mold targets. The number and coverage per unit area are calculated to obtain the "morphological density of mold layer".
[0071] Parameter C (concentric ridge polarization contrast): Using images acquired in different polarization directions, the degree of polarization image is calculated. In the degree of polarization image, the healthy wax layer area is uniform, while the concentric ridge area that is dying will show high degree of polarization stripes. By using Radon transform or directional filtering, the intensity of the directional consistency of these stripes is detected to obtain the "concentric ridge polarization contrast".
[0072] Based on rule-based logical decisions, the processor calls a built-in rule library, which is pre-defined, such as:
[0073] Rule 1: If "Parameter B (morphological density of mold layer on the back)" > threshold T1, it is directly determined as "early blight diagnosis (mid-term and above)" without referring to other parameters;
[0074] Rule 2: If “Parameter B” does not meet the standard, but “Parameter A (early halo index)” > threshold T2 and “Parameter C (concentric fringe polarization contrast)” > threshold T3, then it is judged as “early epidemic suspected (early)”.
[0075] Rule 3: If only "Parameter A" or "Parameter C" slightly exceeds the limit, it is marked as "abnormal, continued observation is recommended";
[0076] The results are output, and the touchscreen displays the judgment result (e.g., "Early Epidemic: Early Stage"), key feature images (e.g., the halo is marked with a red circle, and the mold layer is marked in the microscopic image) and confidence level.
[0077] To comprehensively evaluate the effectiveness of this invention, the following comparative experiments were designed and conducted:
[0078] Early detection capability comparison experiment
[0079] Objective: To verify the detection capability of this invention during the incubation and initial stages of early blight in tomatoes, and to compare it with traditional manual observation and deep learning methods based on visible light images.
[0080] Experimental setup:
[0081] Test materials: 120 tomato seedlings of uniform growth were selected in a controlled greenhouse and randomly divided into 4 groups (30 seedlings in each group). Three of the groups (A, B, and C) were inoculated with a standardized suspension of early blight spores on the third true leaf. Group D served as a blank control (sprayed with sterile water).
[0082] Testing subjects: Starting from the first day after inoculation, the inoculated leaves / corresponding leaf positions of all 120 tomato plants were tested daily.
[0083] Comparison method:
[0084] The method of this invention: The handheld detector described in the embodiments of this invention is used for detection, and the system outputs an "early warning" or a higher level conclusion as "detected".
[0085] Manual observation: Three plant protection technicians with more than 5 years of experience (group information unknown) independently observe the leaves. The appearance of "suspicious water-soaked chlorotic spots" or "extremely small brown spots" on the leaves, as determined by at least two of them, is recorded as "detected".
[0086] Visible light deep learning method: Use a commercial RGB camera to take a picture of the front of the leaf in a standard light source box, input it into a ResNet-50 model that has been pre-trained on a public plant disease dataset and fine-tuned with an additional 10,000 tomato disease images for inference, and record "detection" when the model outputs a confidence score of "early blight" > 0.7.
[0087] Gold standard: Three leaves are randomly selected from each group each day for laboratory tissue isolation and culture and PCR molecular detection to confirm the presence and infection status of pathogens.
[0088] Experimental period: Continued until day 12 after inoculation, or until all methods achieved a 100% detection rate for plants in groups A, B, and C.
[0089] Table 1: Results of the comparative experiment on early detection capabilities
[0090] Days after vaccination (DAI) 1 2 3 4 5 6 7 8 Detection rate of the method of this invention (groups A / B / C) 0% / 0% / 0% 0% / 0% / 0% 23.3% / 0% / 0% 73.3% / 16.7% / 0% 96.7% / 56.7% / 10% 100% / 86.7% / 33% 100% / 100% / 70% 100% / 100% / 96.7% Detection rate by manual observation (Groups A / B / C) 0% / 0% / 0% 0% / 0% / 0% 0% / 0% / 0% 0% / 0% / 0% 13.3% / 0% / 0% 36.7% / 6.7% / 0% 66.7% / 26.7% / 3.3% 90% / 60% / 20% Detection rate of visible light deep learning method (Groups A / B / C) 0% / 0% / 0% 0% / 0% / 0% 0% / 0% / 0% 3.3% / 0% / 0% 10% / 3.3% / 0% 23.3% / 10% / 0% 50% / 16.7% / 3.3% 73.3% / 36.7% / 10% Gold standard for confirming infection rate (Groups A / B / C) 0% / 0% / 0% 10% / 0% / 0% 33.3% / 10% / 0% 86.7% / 30% / 10% 100% / 63.3% / 23.3% 100% / 93.3% / 46.7% 100% / 100% / 80% 100% / 100% / 100%
[0091] Results analysis:
[0092] Significant advantages of ultra-early detection: The method of this invention issues an early warning for some plants in group A (23.3%) as early as the 3rd day after inoculation, which is at least 3-4 days earlier than manual observation and visible light deep learning method. At this time, the gold standard has confirmed that 33.3% of the plants have been infected, but the characteristics that can be distinguished by the naked eye or RGB camera have not yet formed. This is directly due to the capture of the "early water stain halo index", which proves the unique value of this invention in the diagnosis of the incubation period.
[0093] High sensitivity and consistency: During DAI 4-6 days, the detection rate curve of the method of this invention is highly consistent with the infection rate curve confirmed by the gold standard, and it is always ahead of the other two methods; especially at DAI 5, the detection rate of group B has reached 56.7%, while the detection rate of manual and deep learning methods is less than 10%, indicating that the method of this invention has extremely high sensitivity to early disease.
[0094] Limitations and subjectivity of manual observation: Although manual observation can detect symptoms starting on day 5, the progress is slow and fluctuates greatly between different groups, reflecting the problem of strong reliance on experience and inconsistent judgment of atypical symptoms.
[0095] The inherent limitations of visible light deep learning: the method performed poorly throughout the early stages, and its detection rate only increased rapidly on DAI 7-8 days after the lesions became more obvious on the RGB images, confirming its fundamental limitation of not being able to perceive spectral and microscopic morphological changes.
[0096] This invention integrates hyperspectral imaging, leaf underside microscopy, and polarization imaging technologies to directly capture and quantify the specific pathological characteristics of early blight in tomatoes, achieving the following breakthroughs: First, near-infrared spectroscopy analysis enables ultra-early warning of water-soaked lesions during the incubation period, advancing the detection window by 3-5 days. Second, leaf underside microscopy directly captures the diagnostic feature of the pathogen's mold layer, fundamentally solving the problem of confusion with other similar diseases and significantly improving diagnostic specificity. Third, active polarization illumination technology effectively eliminates ambient light interference, ensuring the stability and reliability of the detection results. The entire system is integrated into a portable handheld device, enabling real-time diagnosis in the field and providing reliable technical support for precision prevention and control.
[0097] Finally, it should be noted that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A smart detection method for early blight of tomatoes, characterized in that, Includes the following steps: Simultaneously acquire hyperspectral image information of the front of the tomato leaf, microscopic image information of the back of the leaf, and polarization optical information of the front of the leaf for the same tomato leaf region; From the aforementioned frontal hyperspectral image information, analyze the reflectance characteristics of specific near-infrared bands to extract the early halo index characterizing water-soaked lesions in cells; Based on morphological analysis, the characteristics of black mold were identified from the microscopic images of the leaf back, and the morphological density of the mold layer on the back was calculated. From the polarization optical information on the front side of the blade, the polarization reflection characteristics of the concentric ring pattern structure are analyzed to obtain the polarization contrast of the concentric ring pattern. The early halo index, the morphological density of the mold layer on the back, and the polarization contrast of the concentric lines are input into a set of logical judgment rules based on plant disease knowledge, and the diagnostic conclusions for early blight of tomato and its stages are directly output.
2. The intelligent detection method for early blight of tomatoes according to claim 1, characterized in that, The extraction of the early halo index; specifically: Calculate the reflectance ratio of the blade region in the 780nm and 850nm bands; Identify the closed, low-value ring-shaped regions appearing in the ratio graph; The index value is calculated based on the extent to which the area and ratio of the annular region are below the threshold of the healthy region.
3. The intelligent detection method for early blight of tomatoes according to claim 1, characterized in that, The calculation of the morphological density of the mold layer on the back; include: Threshold segmentation is performed on the microscopic images of the underside of leaves to extract dark connected components within a specific area range; Analyze the circularity and edge texture roughness of each connected region to screen out regions that conform to the morphological characteristics of the early blight fungus mold layer; After statistical screening, the number of regions per unit area and the area ratio are used to generate density parameters.
4. The method according to claim 1, characterized in that, The obtained concentric fringe polarization contrast is specifically: Calculate the degree of polarization image using frontal images obtained from different polarization directions; On the polarization image, directional filtering or Radon transform is used to enhance the radial stripe features, and the strength of the directional consistency of these stripes is measured as a contrast parameter.
5. The method according to claim 1, characterized in that, The logical decision rule set includes at least the following rules: If the density of the mold layer on the back exceeds the first diagnostic threshold, it is directly determined to be the middle to late stage of early blight in tomatoes; If the morphological density of the mold layer on the back does not exceed the first diagnostic threshold, but the early halo index and the polarization contrast of the concentric lines both exceed their respective early warning thresholds, then it is determined to be early stage or suspected early blight of tomato.
6. A tomato early blight intelligent detection system implementing the method of any one of claims 1-5, characterized in that, include: A probe-type acquisition head is used to simultaneously acquire image information of the front and back sides of a blade while it is being held. The upper inner side of the acquisition head is integrated with a hyperspectral-polarization imaging module for acquiring frontal hyperspectral image information and providing polarization illumination. The lower inner side of the acquisition head is integrated with a microscopic camera module for acquiring microscopic image information of the leaf back; A processing and display unit connected to the acquisition head includes a built-in processor for running feature extraction algorithms and logical judgment rule sets, as well as a human-computer interaction interface for displaying diagnostic results.
7. The intelligent detection system for early blight of tomatoes according to claim 6, characterized in that, The hyperspectral-polarization imaging module includes: a hyperspectral imager covering the visible to near-infrared band, and a polarization illumination source consisting of at least three LED rings with different linear polarization directions, wherein the light source provides active illumination with controllable polarization angle for the hyperspectral imager.
8. The intelligent detection system for early blight of tomatoes according to claim 6, characterized in that, The microscope imaging module includes: a fixed-focus microscope lens with a working distance of 1-3 cm, a high-resolution image sensor, and a set of LED supplementary lights to provide lateral illumination for the underside of the leaf.
9. The intelligent detection system for early blight of tomatoes according to claim 6, characterized in that, The inner edge of the clamping head of the probe is equipped with a light-shielding sealing strip to isolate external ambient light interference when clamping the blade.
10. The intelligent detection system for early blight of tomatoes according to claim 6, characterized in that, The processing and display unit is a handheld device, the logic decision rule set is stored in its memory in firmware form, and the processor is an FPGA or an embedded ARM chip.