Bursaphelenchus xylophilus morphological intelligent detection system based on visual identification

Abstract

The invention relates to the technical field of biomorphological analysis, in particular to a pine wood nematode morphological intelligent detection system based on visual identification, which comprises the following modules: a microscopic imaging module used for collecting a multi-focal plane pine wood nematode image sequence and marking focal length and time information; the body fluid refraction response acquisition module is used for acquiring reflection and refraction spectrum information of nematode body fluid in the image sequence and generating body fluid spectrum data; and the contour distortion compensation module is used for establishing a refractive index dynamic model based on the body fluid spectral data, performing dynamic correction on nonlinear contour distortion generated by body fluid refraction in the image sequence data, and generating image data after contour compensation. According to the method, a dynamic body fluid refractive index model is established, and a dynamic contour distortion compensation algorithm is introduced, so that frame-by-frame dynamic correction on nonlinear contour distortion caused by non-uniform distribution of body fluid refraction under the microscopic multi-focal plane imaging condition is realized.

Description

Technical Field

[0001] This invention relates to the field of biological morphology analysis technology, and in particular to a visual recognition-based intelligent morphological detection system for pine wood nematodes. Background Technology

[0002] In existing technologies, microscopic imaging systems do not consider the dynamic optical distortion caused by changes in the refractive index of body fluids with light intensity and internal flow state when extracting the outline of pine wood nematodes. This results in nonlinear bending or local distortion of the nematode outline in the image, which cannot truly reflect the geometric shape of the nematode and thus affects the accuracy of subsequent morphological analysis and classification. Summary of the Invention

[0003] To overcome the above shortcomings, this invention provides a visual recognition-based intelligent morphological detection system for pine wood nematodes, aiming to improve the problem of nematode outline distortion caused by dynamic changes in the refractive index of body fluids.

[0004] In a first aspect, the present invention provides the following technical solution: a visual recognition-based intelligent morphological detection system for pine wood nematodes, comprising the following modules:

[0005] The microscopic imaging module is used to acquire multifocal plane pine wood nematode image sequences and label focal length and time information;

[0006] The body fluid refractive response acquisition module is used to acquire the reflection and refractive spectrum information of nematode body fluid in the image sequence and generate body fluid spectral data;

[0007] The contour distortion compensation module is used to establish a dynamic refractive index model based on body fluid spectral data, dynamically correct the nonlinear contour distortion caused by body fluid refraction in image sequence data, and generate image data after contour compensation.

[0008] The image processing module is used to denoise and register the contour-compensated image data, and extract contour feature data.

[0009] The internal scattering feature recognition module is used to perform internal scattering texture analysis on the contour-compensated image data, extract the scattering intensity and texture direction features after focal plane offset correction, and generate scattering mode feature vectors.

[0010] The contour-scattering coupling analysis module is used to couple contour feature data with scattering pattern feature vectors to form coupled feature data, which is used to distinguish pine wood nematode species.

[0011] By adopting the above technical solution, it is possible to acquire body fluid spectral information in real time and establish a dynamic model of refractive index under microscopic multifocal plane imaging conditions. The nonlinear distortion of the worm's outline caused by body fluid refraction is dynamically corrected frame by frame, thereby accurately restoring the true geometric shape of the worm.

[0012] Preferably, the acquisition of the multifocal plane pine wood nematode image sequence includes:

[0013] Adjust the position of the microscope's focal plane sequentially according to the preset focal length interval and imaging depth;

[0014] Images of pine wood nematode samples were acquired at each focal plane location, and the corresponding focal length information was recorded.

[0015] During continuous acquisition, image acquisition time information is recorded at preset time intervals to generate a time-labeled sequence;

[0016] Preliminary illumination intensity correction and exposure control are performed on the images acquired at each focal plane.

[0017] All focal plane images are serialized and stored according to focal length and time information to form a multi-focal plane image sequence that can be used for subsequent contour compensation and scattering analysis.

[0018] Preferably, the acquisition of reflection and refraction spectral information includes:

[0019] The image sequence of pine wood nematode before contour compensation was used as input data, and the target collection area was selected under microscopic illumination conditions.

[0020] The reflectance and refraction spectra of nematode body fluids in the target area were collected using a spectral detection device.

[0021] The acquired spectral signals are calibrated, including light source intensity correction, background signal suppression, and sensor response correction.

[0022] The calibrated spectral data is correlated with the corresponding focal plane information to form a body fluid spectral data set, and noise filtering and smoothing are performed on the body fluid spectral data.

[0023] Preferably, the establishment of the dynamic refractive index model includes:

[0024] Using body fluid spectral data as input, the spectral signal corresponding to each focal plane is analyzed to extract refractive-related features;

[0025] Based on the refractive characteristics, an initial refractive index model is constructed, and the spectrum-refractive index mapping relationship is established;

[0026] The model is updated over time, and the refractive index parameter is dynamically adjusted based on the image acquisition time and illumination changes.

[0027] The refractive index dynamic model is spatially distributed and optimized by adjusting the refractive index according to the local curvature of the contour and the local structure of the image to reflect the non-uniformity of the nematode body fluid.

[0028] Preferably, dynamic correction of nonlinear contour distortion includes:

[0029] Using the dynamic model of refractive index as input, the initial position and shape information of each focal plane contour in the image sequence are obtained;

[0030] Calculate the refractive distortion vector corresponding to each pixel based on the refractive index dynamic model, and identify local nonlinear contour offsets;

[0031] By combining the local curvature of the contour and the image gradient information, the distorted contour is compensated frame by frame to generate a corrected contour coordinate sequence.

[0032] The corrected contour is smoothed to eliminate high-frequency noise or discontinuities caused by distortion compensation.

[0033] Preferably, the denoising and image registration include:

[0034] The contour-compensated image data is used as input, and spatial denoising is performed on each image, including noise filtering and texture smoothing operations.

[0035] Image registration is performed on multi-focal-plane or multi-time-series images, aligning the images to a unified reference coordinate system through feature matching or contour-based optimization algorithms;

[0036] Contour edge detection is performed on the registered image sequence to extract the insect contour information, and the edges are fitted and smoothed.

[0037] The extracted contour information is transformed into quantifiable contour feature data, including contour length, curvature, width, and local morphological features.

[0038] Preferably, the internal scattering texture analysis includes:

[0039] The contour-compensated image data is used as input, and the analyzable region within the nematode is selected for pixel-level segmentation.

[0040] Within a selected area, perform local grayscale or light intensity distribution analysis on the image to identify changes in the direction and intensity of scattering texture;

[0041] Focal plane offset correction is performed on the scattering texture data, and spatial deviations in texture direction and intensity are corrected by combining multi-focal plane information;

[0042] Feature extraction is performed on the corrected texture data, including scattering intensity statistics, texture direction distribution, and local structure patterns.

[0043] Preferably, the extraction of scattering intensity and texture orientation features includes:

[0044] The image data after internal scattering texture analysis is used as input, and the effective pixels in the analysis area are selected.

[0045] The scattering intensity of the pixels is statistically analyzed, and the local light intensity distribution parameters are calculated, including the average intensity, standard deviation, and local contrast.

[0046] The texture direction of the image region is calculated, and the main texture direction, direction distribution and local direction consistency index are extracted;

[0047] The statistical results of scattering intensity are combined with texture direction features to form a structured feature set, generating feature data that can be used for subsequent analysis or pattern recognition;

[0048] The feature data is standardized to ensure the comparability of features under different focal planes and lighting conditions.

[0049] Preferably, the coupling modeling of contour feature data and scattering pattern feature vectors includes:

[0050] The contour feature data and scattering pattern feature vector are used as input, and matching is performed according to the correspondence between pixel positions or local regions.

[0051] A multidimensional feature matrix is ​​established for the matched data, which includes contour geometric parameters, curvature information, scattering intensity and texture direction features;

[0052] Statistical analysis or machine learning methods are used to fuse the multidimensional feature matrix to generate a coupled feature model that reflects the joint distribution characteristics of contour and internal scattering.

[0053] The coupled feature model is normalized and standardized to form coupled feature data that can be directly used to distinguish pine wood nematode species.

[0054] Output coupled feature data to provide complete multimodal feature input for subsequent classification or recognition.

[0055] The present invention has the following beneficial effects:

[0056] 1. In this invention, by establishing a dynamic model of body fluid refractive index and introducing a dynamic compensation algorithm for contour distortion, the nonlinear distortion of contour caused by non-uniform distribution of body fluid refractive index is dynamically corrected frame by frame under the condition of microscopic multi-focal plane imaging. This solves the problem that the true geometric features of the insect cannot be accurately extracted in traditional microscopic images due to the local bending and morphological distortion of the contour caused by changes in body fluid refractive index.

[0057] 2. In this invention, by introducing spatial registration and focal plane offset correction mechanisms into multi-focal plane image sequences, steady-state reconstruction and unified extraction of the direction and intensity features of internal scattering texture of pine wood nematode are achieved, solving the problem of feature stability caused by the drift of scattering texture position and inconsistency of direction due to small focal plane offset during microscopic imaging.

[0058] 3. In this invention, by performing pixel-level coupling modeling of contour geometric parameters and internal scattering mode feature vectors and constructing a multi-dimensional feature matrix, the joint analysis and fusion modeling of the external morphology and internal scattering features of the nematode are realized. This solves the problem that existing algorithms cannot simultaneously characterize the coupling relationship between contour morphology and internal scattering under dynamic changes in body fluids, thereby achieving accurate differentiation of pine wood nematode species. Attached Figure Description

[0059] Figure 1 This is an architecture diagram of the visual recognition-based intelligent morphological detection system for pine wood nematodes proposed in this invention. Detailed Implementation

[0060] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] In a first embodiment of the present invention, the present invention provides a visual recognition-based intelligent morphological detection system for pine wood nematodes, such as... Figure 1 As shown, it includes the following modules:

[0062] The microscopic imaging module is used to acquire multifocal plane pine wood nematode image sequences and label focal length and time information;

[0063] Furthermore, the acquisition of multifocal plane pine wood nematode image sequences includes:

[0064] Adjust the position of the microscope's focal plane sequentially according to the preset focal length interval and imaging depth;

[0065] Images of pine wood nematode samples were acquired at each focal plane location, and the corresponding focal length information was recorded.

[0066] During continuous acquisition, image acquisition time information is recorded at preset time intervals to generate a time-labeled sequence;

[0067] Preliminary illumination intensity correction and exposure control are performed on the images acquired at each focal plane.

[0068] All focal plane images are serialized and stored according to focal length and time information to form a multi-focal plane image sequence that can be used for subsequent contour compensation and scattering analysis.

[0069] Specifically, the microscopic imaging module is used to acquire multi-focal plane pine wood nematode image sequences and to annotate each image with focal length and time information to ensure that subsequent contour compensation and scattering analysis can use complete spatial and temporal dimension information;

[0070] Before image acquisition, the focal planes are adjusted sequentially by controlling the position of the microscope focal planes according to the preset focal length interval and imaging depth to cover the complete thickness distribution of the nematode sample, ensuring that the images of different focal planes can accurately reflect the structural information of the nematode at different depths; at each focal plane position, the pine wood nematode sample is imaged and the corresponding focal length information is recorded in real time, providing accurate reference data for subsequent focal plane association and three-dimensional contour reconstruction.

[0071] During continuous acquisition, image acquisition time information is recorded at preset time intervals to form a time-labeled sequence. This ensures that the image sequence contains not only spatial information but also temporal information, facilitating the analysis of dynamic changes in body fluids and contour drift. Preliminary illumination intensity correction and exposure control are performed on the images acquired at each focal plane. By adjusting the light source brightness and exposure parameters, the image brightness is made uniform, ensuring the comparability of the images and enabling the subsequent image processing module to stably extract contour features.

[0072] All focal plane images are serialized and stored according to focal length and time information to form a multi-focal plane image sequence. The serialization storage includes the mapping relationship between image data and focal length and time information, ensuring that the subsequent contour distortion compensation module and internal scattering feature recognition module can accurately reference the spatial position and acquisition time of each image. In the specific implementation, the serialization storage can adopt an indexable data structure or database form to enable fast access and retrieval when processing large-scale image sequences.

[0073] Through the above steps, a complete multi-focal plane image sequence can be obtained, ensuring the continuity and traceability of the image sequence in both spatial and temporal dimensions. This provides a sufficient data foundation for contour distortion compensation, scattering texture analysis, and contour-scattering coupling analysis. Simultaneously, it ensures the standardization and systematization of the image acquisition process, facilitating subsequent automated processing and multimodal feature extraction.

[0074] The body fluid refractive response acquisition module is used to acquire the reflection and refractive spectrum information of nematode body fluid in the image sequence and generate body fluid spectral data;

[0075] Furthermore, the acquisition of reflection and refraction spectral information includes:

[0076] The image sequence of pine wood nematode before contour compensation was used as input data, and the target collection area was selected under microscopic illumination conditions.

[0077] The reflectance and refraction spectra of nematode body fluids in the target area were collected using a spectral detection device.

[0078] The acquired spectral signals are calibrated, including light source intensity correction, background signal suppression, and sensor response correction.

[0079] The calibrated spectral data is correlated with the corresponding focal plane information to form a body fluid spectral data set, and noise filtering and smoothing are performed on the body fluid spectral data.

[0080] Specifically, the body fluid refractive response acquisition module is used to acquire the reflection and refractive spectrum information of the body fluid of pine wood nematode in multi-focal plane image sequence, so as to generate complete body fluid spectral data and provide basic data for subsequent contour distortion compensation and refractive index dynamic modeling.

[0081] In the image sequence input stage, the multi-focal plane pine wood nematode image sequence before contour compensation is used as input data, and the target collection area is selected under microscopic illumination. By combining focal plane information and sample location, it is ensured that the collection area can cover the key parts of the nematode's body fluid, which facilitates the acquisition of reflection and refraction spectral characteristics.

[0082] During the specific spectral acquisition process, the body fluid reflectance spectrum R(λ,x,y,z,t) and refractive spectrum T(λ,x,y,z,t) of the target area are acquired by a spectral detection device; where λ represents the wavelength, (x,y,z) represents the three-dimensional spatial coordinates of the pixels in the image sequence, and t represents the acquisition time; the spectral detection device can be equipped with a high-sensitivity sensor and an adjustable light source to ensure the integrity and comparability of the spectral signals;

[0083] The acquired spectral signals are then calibrated, including steps such as light source intensity correction, background signal suppression, and sensor response correction. Spectral calibration can be achieved using the following formula:

[0084]

[0085] Among them, S raw For the raw spectral signal acquired, S bg For the background spectrum, I source R represents the intensity spectrum of the light source. sensor The sensor response function; the calibrated spectral data S calib By associating it with the corresponding focal plane information, a structured set of body fluid spectral data is formed;

[0086] Based on this, noise filtering and smoothing of the body fluid spectral data can be performed using convolution filtering or Gaussian smoothing methods:

[0087] S smooth (λ,x,y,z,t)=S calib (λ,x,y,z,t)*G(σ);

[0088] Where * denotes the convolution operation, G(σ) is the Gaussian kernel function, and σ represents the smoothing parameter, which is used to suppress the influence of random noise on subsequent refractive index modeling;

[0089] The body fluid spectral data set obtained through the above steps can accurately reflect the optical properties of pine wood nematode body fluids at different focal planes and time points, providing sufficient spectral input for contour distortion compensation, refractive index dynamic modeling, and internal scattering feature identification.

[0090] The contour distortion compensation module is used to establish a dynamic refractive index model based on body fluid spectral data, dynamically correct the nonlinear contour distortion caused by body fluid refraction in image sequence data, and generate image data after contour compensation.

[0091] Furthermore, establishing a dynamic model of refractive index includes:

[0092] Using body fluid spectral data as input, the spectral signal corresponding to each focal plane is analyzed to extract refractive-related features;

[0093] Based on the refractive characteristics, an initial refractive index model is constructed, and the spectrum-refractive index mapping relationship is established;

[0094] The model is updated over time, and the refractive index parameter is dynamically adjusted based on the image acquisition time and illumination changes.

[0095] The refractive index dynamic model is spatially distributed and optimized by adjusting the refractive index according to the local curvature of the contour and the local structure of the image to reflect the non-uniformity of the nematode body fluid.

[0096] Furthermore, dynamic correction of nonlinear contour distortion includes:

[0097] Using the dynamic model of refractive index as input, the initial position and shape information of each focal plane contour in the image sequence are obtained;

[0098] Calculate the refractive distortion vector corresponding to each pixel based on the refractive index dynamic model, and identify local nonlinear contour offsets;

[0099] By combining the local curvature of the contour and the image gradient information, the distorted contour is compensated frame by frame to generate a corrected contour coordinate sequence.

[0100] The corrected contour is smoothed to eliminate high-frequency noise or discontinuities caused by distortion compensation.

[0101] Specifically, the contour distortion compensation module is used to establish a dynamic refractive index model based on body fluid spectral data, and to dynamically correct the nonlinear contour distortion caused by the refraction of nematode body fluid in the multi-focal plane image sequence, thereby generating contour-compensated image data. This module takes the body fluid spectral data obtained by the aforementioned body fluid refraction response acquisition module as input, and combines it with the spectral signal corresponding to each focal plane to extract refraction-related features in order to construct an initial refractive index model that reflects the optical properties of body fluid.

[0102] In the specific process of establishing the dynamic refractive index model, the spectral signal corresponding to each focal plane is first analyzed to obtain the fluid refractive-related parameters Δn(λ,x,y,t); where Δn represents the refractive index change, λ represents the wavelength, (x,y) represents the pixel coordinates, and t represents the acquisition time; subsequently, the refractive-related features are mapped to the image spatial coordinates to construct the initial refractive index model n0(x,y,z), establishing the spectrum-refractive index mapping relationship, which can be expressed by the formula:

[0103] n(x,y,z,t)=n0(x,y,z)+f(Δn(λ,x,y,t),I illum (x,y,t));

[0104] Where f represents the combination of spectral characteristics Δn and illumination intensity I. illum A mapping function for dynamically adjusting the refractive index; after the initial model is established, the refractive index parameter is dynamically updated by combining the image acquisition time series and illumination changes to ensure that the model can reflect the non-uniformity and local optical property changes of body fluid at different acquisition times.

[0105] In the spatial distribution optimization stage, the dynamic model of refractive index is locally adjusted. Based on the contour curvature and local image structure information, the refractive index of different regions is corrected to more accurately reflect the non-uniformity of nematode body fluid, while ensuring optical continuity and spatial smoothness.

[0106] When performing dynamic correction of nonlinear contour distortion, the initial position and shape information of each focal plane contour in the image sequence are first obtained, and then the refractive distortion vector corresponding to each pixel is calculated based on the refractive index dynamic model.

[0107]

[0108] in, Δz represents the refractive index gradient, and Δz represents the pixel spacing along the optical axis. Based on the calculated distortion vector, local nonlinear contour offsets are identified, and combined with contour curvature and image gradient information, the distorted contours are compensated frame by frame to generate a corrected contour coordinate sequence. Subsequently, the corrected contours are smoothed, and convolution filtering or Gaussian smoothing methods can be used to eliminate high-frequency noise or discontinuities caused by distortion compensation, ensuring contour continuity and spatial consistency.

[0109] Through the above contour distortion compensation process, a multi-focal plane image sequence corrected by the dynamic refractive index model can be obtained. This sequence accurately reflects the true spatial morphology and contour information of the nematode body fluid, providing reliable basic data for subsequent scattering analysis, morphological recognition and multimodal data fusion, while ensuring the continuity and accuracy of image processing.

[0110] The image processing module is used to denoise and register the contour-compensated image data, and extract contour feature data.

[0111] Furthermore, denoising and image registration include:

[0112] The contour-compensated image data is used as input, and spatial denoising is performed on each image, including noise filtering and texture smoothing operations.

[0113] Image registration is performed on multi-focal-plane or multi-time-series images, aligning the images to a unified reference coordinate system through feature matching or contour-based optimization algorithms;

[0114] Contour edge detection is performed on the registered image sequence to extract the insect contour information, and the edges are fitted and smoothed.

[0115] The extracted contour information is transformed into quantifiable contour feature data, including contour length, curvature, width, and local morphological features.

[0116] Specifically, the image processing module is used to denoise and register the contour-compensated image data, and further extract the nematode contour feature data to provide accurate basic information for subsequent morphological and behavioral analysis.

[0117] In the denoising stage, the contour-compensated image data is input frame by frame into the image processing module. Spatial denoising is performed on each image, including noise filtering and texture smoothing. During the denoising process, algorithms such as convolution filtering, nonlocal mean filtering, or Gaussian filtering can be used to perform weighted smoothing of image pixel values ​​while preserving contour edges and local details. In the texture smoothing stage, local gradient modulation or bilateral filtering is used to suppress the interference of image noise on contour extraction, ensuring the continuity and clarity of contour edges.

[0118] In the image registration stage, multi-focal-plane or multi-time-series images are registered to eliminate spatial misalignment caused by minute displacements or imaging jitter during acquisition. Registration can be based on feature point matching algorithms such as SIFT and ORB, or contour-based optimization methods, by minimizing the contour matching error min∑. i,j ||C i -T(C j )‖ 2 To achieve this, where C i C represents the contour points of the reference image. j The image to be registered is represented by the contour points, and T represents the spatial transformation function, including rotation, translation and affine transformation. Through iterative optimization, all images are aligned to a unified reference coordinate system to ensure that the contour data has spatial consistency.

[0119] In the contour feature extraction stage, contour edge detection is performed on the registered image sequence. The Canny operator, Sobel operator, or an active contour method based on energy minimization can be used to obtain the nematode edge information. After the edge extraction is completed, the contour is fitted and smoothed. For example, the contour points are interpolated by spline curve fitting or polynomial fitting methods to make the contour curve continuous and smooth. Subsequently, the contour geometric information is transformed into quantifiable contour feature data, including contour length, curvature, width, and local morphological features (L,κ,W,M), where L is the contour arc length, κ is the curvature, W is the local width, and M represents the local morphological index, which can be further used for dynamic morphological analysis.

[0120] Through the above image denoising, registration, and contour feature extraction process, continuous, smooth, and quantified nematode contour information can be obtained, ensuring the consistency of data in both time and space dimensions. This provides accurate input data for subsequent body fluid scattering analysis, morphological modeling, and intelligent recognition, while maintaining the integrity of the contour and the authenticity of the morphological features during image processing.

[0121] The internal scattering feature recognition module is used to perform internal scattering texture analysis on the contour-compensated image data, extract the scattering intensity and texture direction features after focal plane offset correction, and generate scattering mode feature vectors.

[0122] Further, internal scattering texture analysis includes:

[0123] The contour-compensated image data is used as input, and the analyzable region within the nematode is selected for pixel-level segmentation.

[0124] Within a selected area, perform local grayscale or light intensity distribution analysis on the image to identify changes in the direction and intensity of scattering texture;

[0125] Focal plane offset correction is performed on the scattering texture data, and spatial deviations in texture direction and intensity are corrected by combining multi-focal plane information;

[0126] Feature extraction is performed on the corrected texture data, including scattering intensity statistics, texture direction distribution, and local structure patterns.

[0127] Furthermore, the extraction of scattering intensity and texture orientation features includes:

[0128] The image data after internal scattering texture analysis is used as input, and the effective pixels in the analysis area are selected.

[0129] The scattering intensity of each pixel is statistically analyzed, and local light intensity distribution parameters are calculated, including average intensity, standard deviation, and local contrast.

[0130] The texture direction of the image region is calculated, and the main texture direction, direction distribution and local direction consistency index are extracted;

[0131] The statistical results of scattering intensity are combined with texture direction features to form a structured feature set, generating feature data that can be used for subsequent analysis or pattern recognition;

[0132] The feature data is standardized to ensure the comparability of features under different focal planes and lighting conditions.

[0133] Specifically, the internal scattering feature recognition module is used to perform pixel-level internal scattering texture analysis on the contour-compensated image data and correct the spatial deviation caused by the focal plane offset in order to extract the scattering intensity and texture direction features after the focal plane offset correction, and then construct a scattering mode feature vector to characterize the internal microstructure.

[0134] In the specific implementation process, the image data after contour compensation is first used as input. Based on the contour positioning information, the analyzable region inside the nematode is selected in the image and divided into pixels or patches to establish pixel-level or block-level analysis units. The division strategy can be based on fixed window, overlapping window or adaptive segmentation method to take into account both spatial resolution and statistical stability. Each analysis unit after division is used as the basic unit for scattering feature extraction in subsequent steps.

[0135] For each analysis unit, local grayscale or light intensity distribution analysis is performed. By calculating the statistical measures of pixel grayscale, the scattering intensity characteristics are grasped. Typical statistical measures include the local average intensity μ. p Local variance With local contrast C p , where it is defined as:

[0136]

[0137] Among them, Ik To analyze the grayscale or light intensity value of the k-th pixel within the cell, N is the total number of pixels in the cell, and ∈ is a small constant to prevent division by zero; these statistics are used to characterize local scattering intensity and brightness fluctuation characteristics.

[0138] To identify texture orientation and its distribution, a gradient operator is introduced to obtain local orientation information of the image, and the pixel-level gradient I is calculated. x with I y And construct a structure tensor J to evaluate directionality. The structure tensor is defined as:

[0139]

[0140] Where <·> represents the local weighted average within the analysis unit; the principal orientation angle θ and the directional consistency index (directional coherence) can be obtained from the structure tensor, and the principal orientation angle can be taken as:

[0141]

[0142] Directional consistency can be characterized by the eigenvalues ​​λ1≥λ2 of the structure tensor, and the directional coherence is defined as:

[0143]

[0144] The value of γ ranges from 0 to 1 and is used to reflect the dominance of the texture direction within the analysis unit; these directional indices are used to describe the directional distribution and local consistency of the scattering texture.

[0145] To address the focal plane offset problem, spatial information from multi-focal plane image sequences is used to correct texture orientation and intensity. The correction strategy includes statistically comparing the intensity and orientation of the same spatial location at different focal planes after multi-focal plane registration to estimate the displacement field Δ(x,y,z,t) caused by the focal plane offset. Furthermore, a reverse mapping correction is performed on the texture orientation and intensity of each analysis unit, ensuring that texture descriptions under different focal planes are comparable on a unified reference plane. Focal plane offset correction can be performed geometrically using an optical axis projection model or a small-angle approximation model. The correction mapping can be expressed as:

[0146] I corr (x,y)=I obs (x+Δ x ,y+Δ y );

[0147] Where I obs For the observed image values, I corr For the corrected image value, Δ x ,Δ y The estimated pixel displacement components; displacement estimation can be obtained by cross-correlation or phase correlation methods between adjacent focal planes of a multi-focal plane image;

[0148] Feature extraction is performed on the texture data corrected for focal plane offset. The feature set includes, but is not limited to, the scattering intensity statistical vector [μ]. p ,σ p C p The system can extract the main texture direction θ, the direction distribution histogram, and the direction coherence γ, and further extract local structural pattern features, such as local second-order statistics or wavelet energy distribution, to characterize the scale and morphological information of the microstructure. To ensure comparability across focal planes and illumination conditions, each feature is standardized. The standardization method can be zero-mean unit variance normalization, that is, the standardized result is calculated for any feature component v.

[0149]

[0150] Where μ v With σ v These are the mean and standard deviation obtained statistically from the training set or the current batch of samples, respectively; after standardization, the features of each unit are concatenated into a structured feature vector in a predetermined order to form a single-frame or temporal scattering pattern feature vector;

[0151] The generation of scattering mode feature vectors includes feature selection, dimensionality reduction and packaging steps. Principal component analysis, linear discriminant analysis or other dimensionality reduction methods can be used to reduce feature dimensionality and enhance discriminability, while retaining the necessary information for subsequent coupled modeling. The generated feature vectors are saved with a unified data structure and metadata (such as the corresponding focal plane index, timestamp and spatial coordinates) so as to perform spatiotemporal correspondence with contour feature data and subsequent fusion processing.

[0152] This embodiment can accurately describe the intensity and direction features of the internal scattering texture of nematodes at the pixel or block scale, and ensure the consistency of cross-focal plane data through focal plane offset correction. This embodiment discloses a complete process from texture direction extraction to directional coherence measurement, from intensity statistics to normalization processing, providing sufficient and traceable scattering feature input for subsequent coupled modeling and species identification.

[0153] The contour-scattering coupling analysis module is used to couple contour feature data with scattering pattern feature vectors to form coupled feature data, which is used to distinguish pine wood nematode species.

[0154] Furthermore, coupling and modeling the contour feature data with the scattering pattern feature vector includes:

[0155] The contour feature data and scattering pattern feature vector are used as input, and matching is performed according to the correspondence between pixel positions or local regions.

[0156] A multidimensional feature matrix is ​​established for the matched data, which includes contour geometric parameters, curvature information, scattering intensity and texture direction features;

[0157] Statistical analysis or machine learning methods are used to fuse the multidimensional feature matrix to generate a coupled feature model that reflects the joint distribution characteristics of contour and internal scattering.

[0158] The coupled feature model is normalized and standardized to form coupled feature data that can be directly used to distinguish pine wood nematode species.

[0159] Output coupled feature data to provide complete multimodal feature input for subsequent classification or recognition.

[0160] Specifically, the contour-scattering coupling analysis module is used to accurately correspond and couple contour feature data with scattering pattern feature vectors in space and time to generate unified coupled feature data that can be used for subsequent recognition or classification. This module achieves the orderly integration and structured expression of multimodal features through pixel-level or local region-level correspondence matching, multidimensional feature matrix construction, fusion modeling, and normalization and standardization processing steps.

[0161] In the specific implementation process, the contour feature data and scattering feature vector are first used as input and spatial correspondence matching is performed. The matching basis includes meta-information such as pixel coordinates, local region center coordinates, focal plane index and timestamp, so as to establish a one-to-one or many-to-one mapping relationship between contour units and scattering units. The matching process can adopt direct mapping based on coordinates, spatial matching based on nearest neighbor search or feature matching based on descriptor similarity, so as to ensure that each contour unit can find the corresponding scattering feature unit, thereby providing a stable corresponding index for subsequent matrix construction.

[0162] A multidimensional feature matrix is ​​constructed from the matched data, with rows representing spatial or local region units and columns representing different feature terms. The matrix elements are composed of contour geometric parameters, local curvature information, contour width and local morphological indices, scattering intensity statistics, main texture direction, directional consistency and other scattering features concatenated in a predetermined order, which can be represented as follows:

[0163]

[0164] Where n represents the number of spatial or local units after matching, m represents the feature dimension of each unit after concatenation, and f ij This represents the j-th feature of the i-th unit. The feature term may include the contour arc length L, local curvature κ, local width W, scattering mean μ, and scattering variance σ. 2 Principal direction angle θ and direction consistency γ, etc.; the matrix construction process also records the focal plane index and timestamp as additional metadata to maintain spatiotemporal correspondence;

[0165] Multidimensional feature matrices are fused to generate coupled feature models. Fusion methods can employ statistical techniques or machine learning frameworks. Statistical methods include principal component analysis, canonical correlation analysis, or covariance analysis, used to reveal the correlation structure between contours and scattering features. Machine learning methods include, but are not limited to, feature fusion based on shallow neural networks, multimodal fusion based on convolutional or fully connected networks, or feature aggregation based on ensemble learning. The fusion steps typically include sub-steps such as feature weighting, mutual information evaluation, dimensionality reduction, and discriminative feature extraction, to generate coupled feature representations that reflect the joint distribution characteristics of contours and internal scattering.

[0166] To ensure consistency in the dimensions and distributions of different features, the coupled features are normalized and standardized. Normalization can be achieved using linear mapping or range scaling, while standardization can be achieved using zero mean and unit variance. Formally, any feature column vector v can be standardized as follows:

[0167]

[0168] in, s is the mean of this feature in the training set or the current batch. v Let be the standard deviation, and ∈ be a small constant to prevent division by zero. The normalized and standardized feature matrices can be further processed for feature selection or dimensionality reduction to generate the final coupled feature data matrix F. coupled And save the corresponding metadata for traceability;

[0169] The coupled feature model can optionally perform model training and adaptive updates. When labeled samples are available, the fusion network or statistical model can be trained under supervision to improve the discrimination ability. At the same time, when new samples are input, the model parameters can be adjusted through incremental learning or online update strategies to adapt to the micro differences between samples and changes in imaging conditions. The model training and update process ensures the reproducibility and verifiability of the results by saving model parameters, feature standardization parameters and mapping relationships.

[0170] The coupled feature data is output in a unified data structure. The output includes the coupled feature matrix, the corresponding spatial cell index, the focal plane and time label, and the mapping metadata for feature interpretation. The output format can be a structured array, a table, or a serialized binary file, so that it can be directly called by subsequent classification, clustering or visualization modules.

[0171] It achieves precise correspondence and orderly fusion of two modal data, contour and scattering, in spatial, temporal and feature dimensions. It discloses a complete technical route from matching strategy, multidimensional matrix construction, fusion modeling to standardized output, ensuring the structure, traceability and interpretability of coupled feature data, and providing sufficient technical support for subsequent automatic identification and discrimination.

[0172] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is 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 visual recognition-based intelligent morphological detection system for pine wood nematodes, characterized in that, Includes the following modules: The microscopic imaging module is used to acquire multifocal plane pine wood nematode image sequences and label focal length and time information; The body fluid refractive response acquisition module is used to acquire the reflection and refractive spectrum information of nematode body fluid in the image sequence and generate body fluid spectral data; The contour distortion compensation module is used to establish a dynamic refractive index model based on body fluid spectral data, dynamically correct the nonlinear contour distortion caused by body fluid refraction in image sequence data, and generate image data after contour compensation. The image processing module is used to denoise and register the contour-compensated image data, and extract contour feature data. The internal scattering feature recognition module is used to perform internal scattering texture analysis on the contour-compensated image data, extract the scattering intensity and texture direction features after focal plane offset correction, and generate scattering mode feature vectors. The contour-scattering coupling analysis module is used to couple contour feature data with scattering pattern feature vectors to form coupled feature data, which is used to distinguish pine wood nematode species.

2. The intelligent morphological detection system for pine wood nematode based on visual recognition according to claim 1, characterized in that, The acquired multifocal plane pine wood nematode image sequence includes: Adjust the position of the microscope's focal plane sequentially according to the preset focal length interval and imaging depth; Images of pine wood nematode samples were acquired at each focal plane location, and the corresponding focal length information was recorded. During continuous acquisition, image acquisition time information is recorded at preset time intervals to generate a time-labeled sequence; Preliminary illumination intensity correction and exposure control are performed on the images acquired at each focal plane. All focal plane images are serialized and stored according to focal length and time information to form a multi-focal plane image sequence that can be used for subsequent contour compensation and scattering analysis.

3. The intelligent morphological detection system for pine wood nematode based on visual recognition according to claim 1, characterized in that, The acquisition of reflection and refraction spectral information includes: The image sequence of pine wood nematode before contour compensation was used as input data, and the target collection area was selected under microscopic illumination conditions. The reflectance and refraction spectra of nematode body fluids in the target area were collected using a spectral detection device. The acquired spectral signals are calibrated, including light source intensity correction, background signal suppression, and sensor response correction. The calibrated spectral data is correlated with the corresponding focal plane information to form a body fluid spectral data set, and noise filtering and smoothing are performed on the body fluid spectral data.

4. The intelligent morphological detection system for pine wood nematode based on visual recognition according to claim 1, characterized in that, The establishment of the dynamic refractive index model includes: Using body fluid spectral data as input, the spectral signal corresponding to each focal plane is analyzed to extract refractive-related features; Based on the refractive characteristics, an initial refractive index model is constructed, and the spectrum-refractive index mapping relationship is established; The model is updated over time, and the refractive index parameter is dynamically adjusted based on the image acquisition time and illumination changes. The refractive index dynamic model is spatially distributed and optimized by adjusting the refractive index according to the local curvature of the contour and the local structure of the image to reflect the non-uniformity of the nematode body fluid.

5. The intelligent morphological detection system for pine wood nematode based on visual recognition according to claim 1, characterized in that, Dynamic correction of nonlinear contour distortion includes: Using the dynamic model of refractive index as input, the initial position and shape information of each focal plane contour in the image sequence are obtained; Calculate the refractive distortion vector corresponding to each pixel based on the refractive index dynamic model, and identify local nonlinear contour offsets; By combining the local curvature of the contour and the image gradient information, the distorted contour is compensated frame by frame to generate a corrected contour coordinate sequence. The corrected contour is smoothed to eliminate high-frequency noise or discontinuities caused by distortion compensation.

6. The intelligent morphological detection system for pine wood nematode based on visual recognition according to claim 1, characterized in that, The denoising and image registration include: The contour-compensated image data is used as input, and spatial denoising is performed on each image, including noise filtering and texture smoothing operations. Image registration is performed on multi-focal-plane or multi-time-series images, aligning the images to a unified reference coordinate system through feature matching or contour-based optimization algorithms; Contour edge detection is performed on the registered image sequence to extract the insect contour information, and the edges are fitted and smoothed. The extracted contour information is transformed into quantifiable contour feature data, including contour length, curvature, width, and local morphological features.

7. The intelligent morphological detection system for pine wood nematode based on visual recognition according to claim 1, characterized in that, The internal scattering texture analysis includes: The contour-compensated image data is used as input, and the analyzable region within the nematode is selected for pixel-level segmentation. Within a selected area, perform local grayscale or light intensity distribution analysis on the image to identify changes in the direction and intensity of scattering texture; Focal plane offset correction is performed on the scattering texture data, and spatial deviations in texture direction and intensity are corrected by combining multi-focal plane information; Feature extraction is performed on the corrected texture data, including scattering intensity statistics, texture direction distribution, and local structure patterns.

8. The intelligent morphological detection system for pine wood nematode based on visual recognition according to claim 1, characterized in that, Extracting scattering intensity and texture orientation features includes: The image data after internal scattering texture analysis is used as input, and the effective pixels in the analysis area are selected. The scattering intensity of the pixels is statistically analyzed, and the local light intensity distribution parameters are calculated, including the average intensity, standard deviation, and local contrast. The texture direction of the image region is calculated, and the main texture direction, direction distribution and local direction consistency index are extracted; The statistical results of scattering intensity are combined with texture direction features to form a structured feature set, generating feature data that can be used for subsequent analysis or pattern recognition; The feature data is standardized to ensure the comparability of features under different focal planes and lighting conditions.

9. The intelligent morphological detection system for pine wood nematode based on visual recognition according to claim 1, characterized in that, The coupling modeling of contour feature data with scattering pattern feature vectors includes: The contour feature data and scattering pattern feature vector are used as input, and matching is performed according to the correspondence between pixel positions or local regions. A multidimensional feature matrix is ​​established for the matched data, which includes contour geometric parameters, curvature information, scattering intensity and texture direction features; Statistical analysis or machine learning methods are used to fuse the multidimensional feature matrix to generate a coupled feature model that reflects the joint distribution characteristics of contour and internal scattering. The coupled feature model is normalized and standardized to form coupled feature data that can be directly used to distinguish pine wood nematode species. Output coupled feature data to provide complete multimodal feature input for subsequent classification or recognition.

Images