Fusion image enhancement intelligent diagnosis method and system for camellia oleifera grafted seedling defects

Abstract

The invention relates to the technical field of agricultural intelligent detection, in particular to a fusion image enhanced intelligent diagnosis method and system for camellia oleifera grafted seedling defects, and the method greatly improves the accuracy of defect detection through integrating adaptive illumination compensation and multi-scale feature fusion technologies. The method comprises the following steps: firstly, carrying out region self-adaptive enhancement processing to eliminate illumination influence, then extracting and fusing multi-scale features from an enhanced image by utilizing a hierarchical feature extraction network and texture-form dual-channel analysis to form comprehensive feature representation, and combining a defect self-attention mechanism to realize 95% of high defect detection rate which is obviously superior to 75% of a traditional method. Besides, through traceability analysis of the knowledge graph, the system can provide defect reasons and suggestions and measures, and through comparison with historical data, an optimization decision and a parameter adjustment scheme are output, so that powerful technical support is provided for intelligent detection and management of the camellia oleifera grafted seedlings.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural intelligent detection, in particular to an intelligent diagnosis method and system for defects of oil-tea camellia grafted seedlings integrating image enhancement, which is applicable to the automatic detection and diagnosis of oil-tea camellia grafted seedlings and the decision-making optimization in the grafting production process. Background Art

[0002] Oil-tea camellia is a unique woody oil tree species in China and is of great significance for the development of forestry economy and the guarantee of edible oil safety. Grafting is an important technique for cultivating oil-tea camellia seedlings, which can effectively improve varieties, promote early fruiting and increase yields. However, various defects often occur in grafted seedlings during the growth process, such as graft union cracking, water shortage and wilting, and disease spot infection. These defects seriously affect the survival rate and growth quality of oil-tea camellia grafted seedlings.

[0003] Traditional defect detection of oil-tea camellia grafted seedlings mainly relies on manual visual inspection, which has problems such as low efficiency, strong subjectivity and high omission detection rate. With the development of computer vision technology, plant detection methods based on machine vision have gradually been applied to agricultural production. For example, Chinese Patent CN102954762B discloses a method and system for measuring external characteristics of grafted seedlings based on machine vision. This method collects images of grafted seedlings from two directions, top view and front view, and extracts external characteristic parameters such as cotyledon parameters, plant height and major and minor axes of the ellipse of seedling diameter of grafted seedlings through image processing algorithms to assist in automatically matching rootstock seedlings and scion seedlings with corresponding seedling diameters during grafting by a grafting machine.

[0004] However, the above-mentioned existing technologies have the following deficiencies:

[0005] 1. Poor light adaptability: Under complex light conditions, the image quality is unstable, affecting the accuracy of feature extraction and measurement;

[0006] 2. Lack of defect diagnosis ability: Only focusing on external feature measurement, it is unable to identify and diagnose the health status or defects of grafted seedlings;

[0007] 3. Using fixed thresholds: Adopting fixed threshold methods such as Otsu is difficult to adapt to grafted seedlings of different varieties and different growth stages;

[0008] 4. Limited feature extraction accuracy: Traditional image processing algorithms have insufficient accuracy in extracting fine features;

[0009] 5. Lack of environmental adaptation mechanism: Unable to adaptively adjust detection parameters according to environmental conditions;

[0010] 6. Lack of intelligent decision-making support: Although it can measure and pair, it lacks intelligent analysis and decision-making support based on historical data.

[0011] Therefore, there is an urgent need to develop a method and system that can accurately identify the defects of oil-tea camellia grafted seedlings in complex environments, conduct intelligent diagnosis, and provide optimized decision-making. Summary of the Invention

[0012] The object of the present invention is to provide an intelligent diagnosis method and system for the defects of oil-tea camellia grafted seedlings integrating image enhancement to solve the above problems existing in the prior art.

[0013] The present invention uses an adaptive light compensation technology to improve the image quality under complex light conditions, realizes the accurate identification of subtle defects through a multi-scale feature fusion technology, realizes the adaptive optimization of system parameters based on an environmental perception dynamic parameter adjustment mechanism, combines an expert knowledge graph to realize the accurate traceability analysis of defects, and provides intelligent decision-making support through historical data mining. This technical solution forms a complete technical closed-loop from image acquisition to intelligent decision-making, and realizes the intelligent diagnosis and analysis of the defects of oil-tea camellia grafted seedlings.

[0014] The present invention proposes an intelligent diagnosis method for the defects of oil-tea camellia grafted seedlings integrating image enhancement, including:

[0015] Obtain the image of the oil-tea camellia grafted seedling and environmental parameters; wherein, the environmental parameters include light intensity, temperature, humidity, and plant moisture;

[0016] Perform adaptive light compensation processing on the image of the oil-tea camellia grafted seedling to obtain an enhanced image; including:

[0017] Based on the environmental parameters, determine the light compensation parameters;

[0018] According to the light compensation parameters, perform regional adaptive enhancement on the image of the oil-tea camellia grafted seedling;

[0019] Extract and fuse multi-scale features from the enhanced image to obtain a feature representation; including:

[0020] Extract multi-scale spatial features from the enhanced image through a hierarchical feature extraction network;

[0021] Extract texture features and morphological features through texture-morphology dual-channel analysis;

[0022] Fuse the multi-scale spatial features, the texture features, and the morphological features to obtain the feature representation;

[0023] Based on the feature representation and the defect self-attention mechanism, identify the defects to obtain a defect diagnosis result;

[0024] Combine the defect diagnosis result with the knowledge graph of the oil-tea camellia grafted seedling for traceability analysis to obtain the defect cause and recommended measures;

[0025] Compare the defect diagnosis results, the defect causes, and the recommended measures with historical data, and output an optimization decision and a parameter adjustment plan.

[0026] Preferably, the adaptive light compensation processing of the oil-tea camellia grafted seedling image specifically includes:

[0027] Divide the oil-tea camellia grafted seedling image into multiple grid regions;

[0028] Calculate the average brightness value of each grid region;

[0029] Use the Gaussian mixture model to identify the brightness abnormal regions and construct an illuminance non-uniformity distribution map;

[0030] Based on the illuminance non-uniformity distribution map, perform illuminance enhancement processing on the weak light regions and illuminance suppression processing on the overexposed regions respectively.

[0031] Preferably, the extraction and fusion of multi-scale features from the enhanced image specifically include:

[0032] Adopt a densely connected convolutional network to extract spatial features of different scales and construct a feature pyramid;

[0033] Use a Gabor filter bank to extract multi-directional and multi-scale texture features;

[0034] Use a multi-morphological operator sequence to extract morphological features;

[0035] Design a weighted adaptive fusion algorithm for feature fusion, where the weight coefficients are dynamically adjusted according to the discriminability of the features.

[0036] Preferably, the defect recognition based on feature representation and defect self-attention mechanism specifically includes:

[0037] Design a spatial self-attention mechanism to capture global dependencies, and the formula is

[0038] According to the attention score matrix A, generate an attention output Z = A·V;

[0039] Locate the defect regions based on the attention output Z;

[0040] Adopt a hierarchical cross-entropy loss function for multi-level classification of defect categories.

[0041] Preferably, the traceability analysis by combining the defect diagnosis results with the oil-tea camellia grafted seedling knowledge graph specifically includes:

[0042] Construct an expert knowledge graph including defect types, environmental factors, grafting operations, variety characteristics, and treatment plans;

[0043] Knowledge reasoning based on attention mechanism graph neural network;

[0044] Use a multi-hop reasoning algorithm to explore the association path between defects and possible causes;

[0045] Calculate the conditional probability P(Cause|Defect, Env) based on the Bayesian network;

[0046] Calculate the causal association strength score, and output the causes and corresponding recommended measures whose scores exceed the threshold.

[0047] Preferably, the comparison of the defect diagnosis results, defect causes, and recommended measures with historical data specifically includes:

[0048] Retrieve similar cases from the multi-dimensional historical data warehouse;

[0049] Construct a multi-objective optimization model: max{w1·survival rate + w2·growth rate + w3·resource efficiency};

[0050] Search for the optimal parameter combination based on the particle swarm algorithm;

[0051] Generate optimization plans for plant pairing, grafting angle, and binding strength.

[0052] Preferably, it further includes an environmental perception dynamic parameter adjustment step:

[0053] Collect environmental parameter data in real time and perform exponentially weighted moving average processing;

[0054] Construct a mapping model between the environmental parameter space Θ and the system performance index space Φ;

[0055] Parameter adjustment algorithm based on performance gradient Adjust the processing parameters;

[0056] Feed back the adjusted processing parameters to the adaptive light compensation and multi-scale feature fusion module.

[0057] Preferably, the defect diagnosis results include:

[0058] Defect types, including normal, graft interface cracking, water shortage wilting, disease spot infection, stem distortion, and malnutrition;

[0059] Defect location coordinates and area;

[0060] Defect severity score, ranging from 0 to 1;

[0061] Defect confidence, with a main category threshold of 0.85 and a sub-category threshold of 0.75.

[0062] Preferably, the hierarchical feature extraction network includes 5 dense connection blocks, each block contains 4 convolutional layers, the growth rate k = 32, forming a feature pyramid structure {64×64×32, 32×32×64, 16×16×128, 8×8×256, 4×4×512}; in the texture-morphology dual-channel analysis, the directions θ of the Gabor filter bank ∈ {0°, 45°, 90°, 135°}, and the scales σ ∈ {1, 2, 4, 8}.

[0063] An intelligent diagnosis system for defects of oil-tea grafted seedlings that performs the fusion image enhancement method of the present invention includes:

[0064] A data acquisition module for acquiring images of oil-tea grafted seedlings and environmental parameters;

[0065] An adaptive light compensation module for performing regional adaptive enhancement on the images of oil-tea grafted seedlings to obtain enhanced images;

[0066] A multi-scale feature fusion module for extracting and fusing multi-scale features from the enhanced images to obtain feature representations;

[0067] A defect recognition module for recognizing defects based on the feature representations and a defect self-attention mechanism to obtain defect diagnosis results;

[0068] A defect traceability module for performing traceability analysis by combining the defect diagnosis results with the knowledge graph of oil-tea grafted seedlings to obtain defect causes and recommended measures;

[0069] A decision optimization module for comparing the defect diagnosis results, the defect causes, and the recommended measures with historical data, and outputting an optimized decision and a parameter adjustment plan;

[0070] An environmental parameter adjustment module for adjusting the system processing parameters in real time to adapt to different environmental conditions;

[0071] Through the collaborative work of the various modules of the system, the intelligent diagnosis and decision optimization of defects of oil-tea grafted seedlings are realized.

[0072] The present invention has the following beneficial effects:

[0073] 1. Improve the accuracy of defect detection: Through adaptive light compensation and multi-scale feature fusion, the defect detection rate reaches 95%, which is much higher than 75% of traditional methods;

[0074] 2. Enhance the environmental adaptability: Based on environmental perception and dynamic parameter adjustment, stable effects can still be maintained under the condition of ±70% light change;

[0075] 3. Achieve precise defect diagnosis: It can identify tiny defects (≥0.5mm) with an accuracy rate of 92%, and provide detailed cause analysis and solutions;

[0076] 4. Optimize production decisions: The grafting survival rate is increased by 15%, manual intervention is reduced by 65%, and production efficiency is improved by 30%;

[0077] 5. System self-optimization: As the usage time increases, the system performance continues to improve, forming a virtuous cycle. Brief Description of the Drawings

[0078] Figure 1 is the overall flowchart of the intelligent defect diagnosis method for oil-tea camellia grafted seedlings integrating image enhancement of the present invention;

[0079] Figure 2 is the flowchart of the adaptive light compensation processing of the present invention;

[0080] Figure 3 is the flowchart of the multi-scale feature fusion of the present invention;

[0081] Figure 4 is the structural schematic diagram of the defect self-attention mechanism of the present invention;

[0082] Figure 5 is the flowchart of the defect traceability analysis driven by the knowledge graph of the present invention;

[0083] Figure 6 is the flowchart of the environmental perception dynamic parameter adjustment mechanism of the present invention;

[0084] Figure 7 is the architecture diagram of the intelligent defect diagnosis system for oil-tea camellia grafted seedlings integrating image enhancement of the present invention;

[0085] Figure 8 is the structural diagram of the adaptive light compensation module of the present invention;

[0086] Figure 9 is the structural diagram of the multi-scale feature fusion module of the present invention;

[0087] Figure 10 is the data flow relationship diagram between the modules of the system of the present invention. Detailed Embodiments

[0088] Please refer to the attached Figures 1-10 , and the present invention will be further described in detail below with reference to the drawings and embodiments.

[0089] Embodiment 1: Intelligent Defect Diagnosis Method for Oil-tea Camellia Grafted Seedlings Integrating Image Enhancement

[0090] As Figure 1As shown in the figure, the intelligent defect diagnosis method for oil-tea camellia grafted seedlings with fused image enhancement provided by the present invention includes the following steps:

[0091] Step S1: Obtain the image of the oil-tea camellia grafted seedling and environmental parameters;

[0092] Step S2: Perform adaptive light compensation processing on the image of the oil-tea camellia grafted seedling to obtain an enhanced image;

[0093] Step S3: Extract and fuse multi-scale features from the enhanced image to obtain a feature representation;

[0094] Step S4: Identify defects based on the feature representation and the defect self-attention mechanism to obtain a defect diagnosis result;

[0095] Step S5: Combine the defect diagnosis result with the knowledge graph of the oil-tea camellia grafted seedling for traceability analysis to obtain the defect cause and recommended measures;

[0096] Step S6: Compare the defect diagnosis result, the defect cause, and the recommended measures with historical data, and output an optimization decision and a parameter adjustment plan.

[0097] In step S1, the image of the oil-tea camellia grafted seedling and environmental parameters are obtained. Among them, the environmental parameters include light intensity, temperature, humidity, and plant moisture. The image of the oil-tea camellia grafted seedling can be obtained by a high-precision RGB camera, and the environmental parameters are collected by a corresponding sensor array. These data together constitute the basic input for subsequent processing.

[0098] In step S2, adaptive light compensation processing is performed on the image of the oil-tea camellia grafted seedling to obtain an enhanced image. First, the light compensation parameters are determined based on the environmental parameters; then, based on the light compensation parameters, regional adaptive enhancement is performed on the image of the oil-tea camellia grafted seedling. This step can effectively solve the problem of unstable image quality in a complex light environment and provide a high-quality image basis for subsequent feature extraction.

[0099] In step S3, multi-scale features are extracted from the enhanced image and fused to obtain a feature representation. This step includes: extracting multi-scale spatial features from the enhanced image through a hierarchical feature extraction network; extracting texture features and morphological features through texture-morphology dual-channel analysis; fusing the multi-scale spatial features, texture features, and morphological features to obtain a feature representation. This multi-dimensional feature fusion can effectively capture various subtle features of the oil-tea camellia grafted seedling and provide a comprehensive information basis for defect identification.

[0100] In step S4, defects are identified based on the feature representation and the defect self-attention mechanism to obtain a defect diagnosis result. The self-attention mechanism can capture the global dependence relationship between features and effectively improve the accuracy and robustness of defect identification. The defect diagnosis result includes information such as defect type, defect location, and severity.

[0101] In step S5, the defect diagnosis results are combined with the knowledge graph of oil-tea camellia grafted seedlings for traceability analysis to obtain the defect causes and recommended measures. The knowledge graph contains rich expert experience and domain knowledge. Through the graph inference engine, traceability analysis from the defect phenomenon to the possible causes can be realized, and corresponding recommended measures can be given.

[0102] In step S6, the defect diagnosis results, defect causes, and recommended measures are compared with historical data, and optimized decisions and parameter adjustment plans are output. By analyzing the treatment effects of similar cases in historical data and combining with the current situation, the system generates the optimal decision plan and parameter adjustment suggestions to improve the grafting survival rate and production efficiency.

[0103] Through the above six key steps, this method forms a complete technical closed-loop from image acquisition to intelligent decision-making. Each step is closely connected, the data flow is clear, and the intelligent diagnosis and analysis of the defects of oil-tea camellia grafted seedlings are realized.

[0104] Embodiment 2: Implementation method of adaptive light compensation processing

[0105] As Figure 2 shown, this embodiment details the adaptive light compensation processing in step S2, which specifically includes:

[0106] First, the oil-tea camellia grafted seedling image is divided into multiple grid regions. Preferably, the image is evenly divided into 25×25 grids. The size of each grid region is appropriate, which can capture local light changes and will not be too fragmented to increase the computational burden.

[0107] Then, the average brightness value of each grid region is calculated. For an RGB image, it can be converted to the HSV color space, and the average value of the V channel (brightness channel) is calculated as the brightness feature of this region. This method can reflect the brightness perceived by the human eye more than directly calculating in the RGB space.

[0108] Next, the Gaussian mixture model is used to identify the regions with abnormal brightness and construct an illuminance non-uniformity distribution map. The Gaussian mixture model can be expressed as:

[0109]

[0110] where λ i is the mixing coefficient of the i-th Gaussian component, satisfying is the probability density function of the i-th Gaussian component, μ i is the mean value, ∑ i is the covariance matrix; K is the number of Gaussian components, preferably set to 3, corresponding to the normal brightness, over-dark, and over-bright regions respectively.

[0111] Finally, based on the illuminance non-uniformity distribution map, illuminance enhancement processing is respectively performed on the low-light regions and illuminance suppression processing is performed on the overexposed regions. For the low-light regions, an illuminance enhancement coefficient η = 1 + (Imax - I) / Imax × δ is designed, where I is the current luminance value of the region, Imax is the maximum luminance value in the image, and δ is the enhancement factor, and the preferred value range is [0.5, 1.5]; for the overexposed regions, an illuminance suppression coefficient ω = 1 - (I - Imin) / Imax × ε is designed, where Imin is the minimum luminance value in the image and ε is the suppression factor, and the preferred value range is [0.3, 0.8].

[0112] Through this region-adaptive light compensation processing, the problem of unstable image quality under complex lighting conditions can be effectively solved, providing a high-quality image basis for subsequent feature extraction and defect recognition. In practical applications, compared with the traditional global histogram equalization, this method can more effectively enhance the detail regions while maintaining the overall naturalness of the image, improving the accuracy of subsequent processing.

[0113] Example 3: Implementation method of multi-scale feature fusion

[0114] As Figure 3 shown, this example details the implementation method of multi-scale feature fusion in step S3, specifically including:

[0115] First, a dense connection convolutional network is used to extract spatial features of different scales and construct a feature pyramid. In this example, the dense connection network contains 5 dense connection blocks, each block contains 4 convolutional layers, and the growth rate k = 32. The feature pyramid structure is designed as {64×64×32, 32×32×64, 16×16×128, 8×8×256, 4×4×512}, which can capture multi-scale information from details to the global. The dense connection design enables a direct connection between any two layers in the network, effectively alleviating the problem of gradient disappearance and improving the efficiency of feature extraction.

[0116] Then, a Gabor filter bank is used to extract multi-directional and multi-scale texture features. The Gabor filter is a band-pass filter that simulates the perception mechanism of the human visual system and has a strong response to textures of specific directions and scales. In this example, a Gabor filter bank with directions θ ∈ {0°, 45°, 90°, 135°} and scales δ ∈ {1, 2, 4, 8} is designed to filter the enhanced image, obtaining 16 response maps, which reflect the texture information of different directions and scales in the image.

[0117] Next, a multi-morphological operator sequence is used to extract morphological features. Morphological features can reflect the geometric structure and topological characteristics of the surface of oil-tea camellia grafted seedlings, and are particularly effective for detecting defects such as cracks and folds. In this embodiment, a multi-morphological operator sequence including opening operation, closing operation, top-hat transformation, and bottom-hat transformation is designed, and the image is processed using structural elements of different sizes and shapes to extract morphological features.

[0118] Finally, a weighted adaptive fusion algorithm is designed for feature fusion, where the weight coefficients are dynamically adjusted according to the discriminability of the features. For each feature vector f i , calculate its discriminability score s i :

[0119]

[0120] where, σ between (f i ) represents the between-class variance, and σ within (f i ) represents the within-class variance. The higher the discriminability score, the more effective the feature is in distinguishing different classes. Based on the discriminability score, calculate the weight w i of each feature:

[0121]

[0122] Finally, the feature obtained by weighted fusion is expressed as:

[0123]

[0124] Through this multi-scale and multi-type feature fusion method, various features of oil-tea camellia grafted seedlings can be comprehensively captured, providing a rich information basis for subsequent defect recognition. Compared with single features, the fused features have stronger expressive ability and robustness, can adapt to complex and changeable detection environments, and improve the accuracy of defect recognition.

[0125] Example 4: Defect Recognition Method Based on Defect Self-Attention Mechanism

[0126] As Figure 4 shown, this embodiment details the method for recognizing defects based on feature representation and defect self-attention mechanism in step S4, specifically including:

[0127] First, a spatial self-attention mechanism is designed to capture global dependencies. The core idea of the self-attention mechanism is to calculate the correlation between any two positions in the feature map, enabling the model to pay attention to other positions related to the current position, thereby capturing the global dependencies of the features. Mathematically, the self-attention mechanism can be expressed as:

[0128]

[0129] Among them, Q, K, and V represent the query matrix, key matrix, and value matrix respectively, all of which are obtained by different linear transformations of the input feature X; d is the feature dimension; the softmax function is used to normalize the attention scores into a probability distribution.

[0130] Then, according to the attention score matrix A, the attention output Z is generated:

[0131] Z = A · V,

[0132] The attention output Z integrates global dependencies and can better express the association between the defect region and its context.

[0133] Next, based on the attention output Z, the defect region is located. By setting an appropriate threshold, the regions with higher attention scores can be identified as potential defect regions. On this basis, combined with morphological processing and connected component analysis, the position and range of the defects are further refined.

[0134] Finally, a hierarchical cross-entropy loss function is used for multi-level classification of defect categories. The hierarchical cross-entropy loss function takes into account the hierarchical relationship between defect categories and can better handle the class imbalance problem:

[0135] L = -

[0136] where y i is the class label, p i is the predicted probability, λ is the trade-off factor, and mask(x i ) is a masking function used to select specific samples to participate in the hierarchical loss calculation.

[0137] Through the above method, the system can accurately identify various defects of oil-tea camellia grafted seedlings, including grafting interface cracking, water shortage and wilting, lesion infection, stem distortion, and malnutrition, etc. The defect diagnosis results include not only the defect type, but also information such as the location, area, severity, and confidence of the defect, providing a basis for subsequent traceability analysis and decision-making optimization.

[0138] Preferably, the main category confidence threshold is set to 0.85, and the sub-category confidence threshold is set to 0.75 to balance the detection precision and recall rate. Experiments show that under the condition of ±70% change in light conditions, the defect detection rate of this method can still remain above 95%, far exceeding the performance of traditional methods.

[0139] Example 5: Knowledge Graph-Driven Defect Traceability Analysis Method

[0140] As Figure 5As shown, this embodiment details the method of traceability analysis by combining the defect diagnosis results with the knowledge graph of oil-tea camellia grafted seedlings in step S5, specifically including:

[0141] First, construct an expert knowledge graph that includes defect types, environmental factors, grafting operations, variety characteristics, and treatment plans. A knowledge graph is a structured form of knowledge representation, consisting of entities, relationships, and attributes, and can describe the concepts, classifications, and interrelationships of domain knowledge. In this embodiment, the entity types of the knowledge graph include defect types, environmental factors, grafting operations, variety characteristics, and treatment plans; the relationship types include cause, affect, associate, inhibit, and promote. Knowledge is stored in the form of triples, such as <high temperature and dryness, cause, graft interface cracking>, forming a rich and structured knowledge base.

[0142] Then, perform knowledge reasoning based on a graph neural network with an attention mechanism. A graph neural network is a type of deep learning model specialized for processing graph-structured data, which can learn the representations of nodes and edges and perform complex reasoning tasks. In this embodiment, the TransE algorithm is used for knowledge representation learning:

[0143] ∥h+r-t∥2,

[0144] where h, r, and t are the vector representations of the head entity, relationship, and tail entity respectively. The core idea of this algorithm is that if (h, r, t) is a valid triple, then h + r should be close to t.

[0145] Next, use a multi-hop reasoning algorithm to explore the association paths between defects and possible causes. Multi-hop reasoning refers to connecting entities that are not directly related through multiple relationship paths to discover potential associations. In this embodiment, the maximum reasoning depth d = 5 and the threshold τR = 0.65 are set, and by searching for paths starting from the defect node in the knowledge graph, possible cause nodes are found.

[0146] Then, calculate the conditional probability P(Cause|Defect, Env) based on a Bayesian network, that is, the probability of various possible causes given the defect type and environmental conditions. A Bayesian network is a probabilistic graphical model that can represent the conditional dependence relationships between variables and answer various queries through probabilistic reasoning.

[0147] Finally, calculate the causal association strength score, and output the causes and corresponding recommended measures whose scores exceed the threshold. The calculation formula for the causal association strength score is:

[0148] S = α·P + β·F + γ·R,

[0149] Among them, P is the conditional probability value, F is the historical frequency, R is the expert rule score, and α, β, and γ are weight coefficients, satisfying α + β + γ = 1. Preferably, α = 0.5, β = 0.3, and γ = 0.2 are set to balance the roles of probability reasoning, historical statistics, and expert knowledge. The scoring threshold τS is set to 0.7, and causal relationships below this value are filtered out.

[0150] Through defect traceability analysis driven by a knowledge graph, the system can associate defect phenomena with possible causes, provide more comprehensive and in-depth diagnostic results, and provide a scientific basis for subsequent governance measures. Compared with pure data-driven methods, this method combining expert knowledge has obvious advantages in dealing with rare defects and complex causal relationships and can better utilize the experience and knowledge of domain experts.

[0151] Example 6: Historical data comparison and optimization decision-making method

[0152] This example details the method of comparing defect diagnosis results, defect causes, and recommended measures with historical data in step S6 and outputting an optimization decision and parameter adjustment plan, which specifically includes:

[0153] First, retrieve similar cases from the multi-dimensional historical data warehouse. The historical data warehouse is divided into an original data layer, a statistical summary layer, and an analysis and prediction layer according to the data level; it is divided into production batch data, environmental data, operation records, test results, and survival rate data according to the data category; and it is divided into hourly, daily, weekly, monthly, quarterly, and annual levels according to the time granularity. The retrieval of similar cases is based on a multi-feature matching algorithm, considering the similarity of factors such as defect type, severity, and environmental conditions, and finding the cases in history that are most similar to the current situation.

[0154] Then, construct a multi-objective optimization model:

[0155] max{w1·survival rate + w2·growth rate + w3·resource efficiency},

[0156] Among them, w1, w2, and w3 are weight coefficients, satisfying w1 + w2 + w3 = 1. These weight coefficients can be adjusted according to different emphases of production requirements. For example, in an urgent production task, the weight of the survival rate can be increased; in long-term cultivation, the weight of the growth rate can be increased; in regular production, the weights of the three can be balanced.

[0157] Next, the particle swarm optimization algorithm is used to search for the optimal parameter combination. The particle swarm optimization algorithm is a swarm intelligence optimization algorithm that finds the optimal solution by simulating the foraging behavior of a bird flock. In this embodiment, the number of particles N = 50, the maximum number of iterations T = 200, and the parameter combination search space includes plant pairing {d1 / d2 ∈ [0.8, 1.2]}, grafting angle {θ ∈ [15°, 45°]}, and binding strength {F ∈ [0.5N, 2N]}. Each particle represents a set of parameter combinations, and its fitness is determined by the objective function value of the multi-objective optimization model. By iteratively updating the position and velocity of the particles, the optimal parameter combination is finally found.

[0158] Finally, an optimized plan for plant pairing, grafting angle, and binding strength is generated. The optimized plan not only includes the optimal values of the parameters but also includes the reasonable ranges of the parameters, expected effects, and applicable conditions, so that the operators can better understand and execute. In addition, the system will also give a survival rate prediction and possible risk warnings based on the analysis results of historical data to help production managers make more informed decisions.

[0159] Through historical data comparison and optimization decision-making, the system can combine the current diagnosis results with historical experience to provide more scientific and practical decision-making suggestions, effectively improving the survival rate and production efficiency of oil-tea camellia grafted seedlings. Practice has shown that this method can increase the grafting survival rate by 15%, reduce manual intervention by 65%, and improve production efficiency by 30%, which is of great significance for improving the modernization level of the oil-tea camellia industry.

[0160] Example 7: Environmental perception dynamic parameter adjustment method

[0161] As Figure 6 shown, this embodiment details the environmental perception dynamic parameter adjustment method in claim 7. This method is a supplement to the above six main steps and runs through the entire diagnosis process to ensure the adaptability and stability of the system under different environmental conditions. Specifically, it includes:

[0162] First, environmental parameter data is collected in real time and processed using exponential weighted moving average. Exponential weighted moving average (EWMA) is a commonly used time series smoothing method, and its calculation formula is:

[0163] S t = α·X t +(1 - α)·S t-1 ,

[0164] where S t is the current smoothed value, X t is the current observed value, S t-1is the smoothed value at the previous moment, α is the smoothing coefficient, and its value range is (0, 1). The larger α is, the higher the weight of the new data, and the smoother curve is closer to the original data; the smaller α is, the higher the weight of the historical data, and the more obvious the smoothing effect. In this embodiment, α is set to 0.3 to balance real-time response and stability.

[0165] Then, a mapping model between the environmental parameter space Θ and the system performance index space Φ is constructed. The environmental parameter space is a multi-dimensional space, including dimensions such as light intensity, temperature, humidity, and plant moisture; the system performance index space is also a multi-dimensional space, including dimensions such as detection accuracy, processing speed, and resource consumption. The mapping model f: Θ → Φ is a function from the environmental parameter space to the system performance index space, which describes the impact of environmental conditions on system performance. In this embodiment, a neural network is used to construct the mapping model, and the network structure is: input layer (environmental parameter vector) → hidden layer 1 (128) → hidden layer 2 (64) → output layer (performance index vector).

[0166] Next, the processing parameters are adjusted based on the parameter adjustment algorithm of the performance gradient. The goal of parameter adjustment is to maximize the system performance index Φ(θ), where θ is the system processing parameter. The adjustment formula is:

[0167]

[0168] where, θ t is the current parameter, θ t+1 is the adjusted parameter, η is the learning rate, is the gradient of the performance index with respect to the parameter. To prevent drastic fluctuations of the parameter, a parameter change range constraint is set: |θ t+1 - θ t | ≤ Δmax. At the same time, the learning rate can also be adaptively adjusted:

[0169] η t+1 = η t ·(1 + δ · sign(Φ(θ t+1 )) - Φ(θ t ))),

[0170] where, δ is the adjustment step size, and the sign function takes the value of 1 or -1 according to the direction of performance change.

[0171] Finally, the adjusted processing parameters are fed back to the adaptive light compensation and multi-scale feature fusion module. Specifically, the processing parameters include contrast parameters, threshold parameters, filtering parameters, and attention mechanism parameters, etc., which respectively affect image enhancement, feature extraction, defect recognition, etc. By dynamically adjusting these parameters, the system can better adapt to different environmental conditions and maintain stable performance.

[0172] The environmental perception dynamic parameter adjustment mechanism is one of the important innovations of the present invention. It enables the system to have self - adaptation ability and can operate efficiently and stably in complex and changeable environments. Experiments show that when environmental conditions change frequently, the performance of the system with this mechanism decreases by no more than 8%, while the performance of the traditional fixed - parameter system decreases by up to 35%, which reflects the significant advantages of the present invention.

[0173] Example 8: Intelligent Defect Diagnosis System for Camellia oleifera Grafted Seedlings Integrated with Image Enhancement

[0174] As Figure 7 shown, the present invention also provides an intelligent defect diagnosis system for Camellia oleifera grafted seedlings integrated with image enhancement, which is used to implement the above - mentioned method and includes:

[0175] A data acquisition module 1, which is used to obtain images of Camellia oleifera grafted seedlings and environmental parameters;

[0176] An adaptive light compensation module 2, which is used to perform region - adaptive enhancement on the images of Camellia oleifera grafted seedlings to obtain enhanced images;

[0177] A multi - scale feature fusion module 3, which is used to extract and fuse multi - scale features from the enhanced images to obtain feature representations;

[0178] A defect recognition module 4, which is used to recognize defects based on the feature representations and the defect self - attention mechanism to obtain defect diagnosis results;

[0179] A defect traceability module 5, which is used to perform traceability analysis by combining the defect diagnosis results with the knowledge graph of Camellia oleifera grafted seedlings to obtain defect causes and recommended measures;

[0180] A decision - making optimization module 6, which is used to compare the defect diagnosis results, defect causes and recommended measures with historical data and output optimized decisions and parameter adjustment plans;

[0181] An environmental parameter adjustment module 7, which is used to adjust the system processing parameters in real - time to adapt to different environmental conditions.

[0182] As Figure 10As shown in the figure, there is a clear data flow relationship among the system modules. The images of oil-tea camellia grafted seedlings and environmental parameters collected by the data acquisition module 1 are respectively input into the adaptive light compensation module 2 and the environmental parameter adjustment module 7. The output of the adaptive light compensation module 2, i.e., the enhanced image, is used as the input of the multi-scale feature fusion module 3. The output of the multi-scale feature fusion module 3, i.e., the feature representation, is used as the input of the defect recognition module 4. The output of the defect recognition module 4, i.e., the defect diagnosis result, is used as the input of the defect traceability module 5. The output of the defect traceability module 5, i.e., the defect cause and suggested measures, together with the defect diagnosis result, are input into the decision optimization module 6. The output of the environmental parameter adjustment module 7, i.e., the adjusted processing parameters, are fed back to the adaptive light compensation module 2, the multi-scale feature fusion module 3, and the defect recognition module 4 to form a closed-loop control.

[0183] Through this modular design and clear data flow relationship, the system can operate efficiently and stably. Each module cooperates with each other to jointly achieve the intelligent diagnosis and decision optimization of the defects of oil-tea camellia grafted seedlings.

[0184] Example 9: Hardware Structure and Software Implementation of the System

[0185] This example details the hardware structure and software implementation plan of the system to ensure its industrial practicability and feasibility of implementation.

[0186] In terms of hardware, the system includes:

[0187] 1. Image acquisition unit: It includes an industrial-grade RGB camera (resolution ≥ 1920×1080, frame rate ≥ 30fps) and a supplementary light device, which are used to obtain high-quality images of oil-tea camellia grafted seedlings;

[0188] 2. Environmental sensing unit: It includes a light intensity sensor, a temperature sensor, a humidity sensor, and a plant moisture sensor, which are used to monitor environmental parameters in real time;

[0189] 3. Computing and processing unit: It includes an industrial-grade edge computing device, configured with a 4-core CPU (≥ 2.5GHz), 8GB RAM, and supports GPU acceleration, which is used to execute computationally intensive tasks such as image processing, feature extraction, and defect recognition;

[0190] 4. Storage unit: It includes an SSD (≥ 256GB) for storing system software and real-time data, and an HDD (≥ 2TB) for storing historical data;

[0191] 5. Human-computer interaction unit: It includes a high-resolution touch display screen and an industrial-grade control terminal, which are used to display the diagnosis results and receive user instructions;

[0192] 6. Communication unit: It includes a wired network interface and a wireless communication module, which support local network and remote connections.

[0193] In terms of software, the system adopts a layered architecture design:

[0194] 1. Bottom - layer driver layer: Responsible for the control and management of hardware devices, including camera drivers, sensor drivers, communication drivers, etc.;

[0195] 2. Data processing layer: Responsible for data acquisition, pre - processing and storage, including image capture, environmental parameter acquisition, and data persistence, etc.;

[0196] 3. Algorithm core layer: Responsible for implementing core algorithms, including adaptive light compensation, multi - scale feature fusion, defect self - attention mechanism, knowledge graph reasoning, and decision optimization, etc.;

[0197] 4. Application service layer: Responsible for providing business functions, including defect diagnosis, cause analysis, decision - making suggestions, and system management, etc.;

[0198] 5. User interface layer: Responsible for human - machine interaction, including visual display, operation control, and alarm notification, etc.

[0199] The software implementation of the system is based on the following technology stack:

[0200] 1. Operating system: Linux real - time version (RT - Linux) or Windows Embedded Industrial Edition;

[0201] 2. Development framework: OpenCV (≥4.5.0) for image processing, TensorFlow Lite (≥2.5.0) for running AI models;

[0202] 3. Database: Redis for real - time data, PostgreSQL for historical data;

[0203] 4. Communication protocol: OPC UA protocol for device - to - device communication, MQTT protocol for remote transmission;

[0204] 5. User interface: Developed based on the Web technology stack (HTML5 + CSS3 + JavaScript), supporting responsive design and adapting to different devices.

[0205] Through reasonable hardware configuration and software architecture design, the system has good industrial practicability and feasibility, can run stably in the actual production environment, and provides reliable support for the intelligent diagnosis of oil - tea grafted seedlings.

[0206] Example 10: Analysis of System Application Cases

[0207] This example demonstrates the actual operation effect and technical advantages of the system through a specific application case.

[0208] A certain oil-tea camellia planting base uses this system to conduct defect diagnosis on a batch of oil-tea camellia grafted seedlings during the grafting season. The specific process is as follows:

[0209] First, the data acquisition module obtains images of the oil-tea camellia grafted seedlings and records the environmental parameters simultaneously: light intensity 850 lux, temperature 26 °C, humidity 65%, and plant water content 72%.

[0210] Then, the adaptive light compensation module processes the original image. Since the light conditions were uneven on the test day, there were obvious shadow areas in the original image, which affected feature extraction. The system divided the image into 25×25 grids, identified the regions with abnormal brightness, enhanced the weak light regions and suppressed the overexposed regions specifically, and obtained a balanced and clear enhanced image.

[0211] Next, the multi-scale feature fusion module extracts features from the enhanced image. The system extracts multi-scale spatial features through a densely connected convolutional network, texture features through a Gabor filter bank, and morphological features through a sequence of multi-morphological operators, and fuses these features with weights to obtain a comprehensive feature representation.

[0212] Then, the defect recognition module recognizes defects based on the feature representation and self-attention mechanism. The system detects a tiny crack at the grafting interface of an oil-tea camellia grafted seedling. The defect type is grafting interface cracking, the confidence level is 0.92, and the severity is 0.4 (medium).

[0213] Next, the defect tracing module conducts analysis in combination with the knowledge graph. The system queries the causal path related to grafting interface cracking in the knowledge graph, combines the current environmental parameters and historical data, infers that the possible causes are loose binding (probability 0.75) and dry environment (probability 0.62), and gives corresponding recommended measures: appropriately increase the binding force (recommended value: 1.5 N) and increase the environmental humidity (recommended value: 75%).

[0214] Finally, the decision optimization module generates an optimized decision according to the defect diagnosis results and tracing analysis, in combination with historical data. The system recommends a comprehensive inspection of the binding conditions of the grafted seedlings of the same batch and adjustment of the grafting parameters: the diameter ratio of the rootstock / scion is adjusted to 1.1 (originally 1.0), the grafting angle is adjusted to 30° (originally 25°), and the binding force is increased by 25%. At the same time, the system predicts that the survival rate after these adjustments can be increased by about 12%.

[0215] This case demonstrates the operation process and effects of the system in practical applications. Through the adaptive light compensation technology, the system solves the problem of unstable image quality under complex lighting conditions; through the multi-scale feature fusion technology, the system achieves precise recognition of subtle defects; through the knowledge graph-driven traceability analysis, the system provides professional cause diagnosis and recommended measures; through the mining and analysis of historical data, the system generates a scientific decision optimization plan. The whole process reflects the collaborative work among the system modules, forming a complete technical closed-loop from input to output, and realizing the intelligent diagnosis and decision optimization of the defects of Camellia oleifera grafted seedlings.

[0216] The intelligent diagnosis method and system for the defects of Camellia oleifera grafted seedlings integrating image enhancement of the present invention have important industrial application values:

[0217] 1. For Camellia oleifera planting enterprises, this system can significantly improve the quality control level of grafted seedlings, reduce the manual detection cost, and improve the production efficiency;

[0218] 2. For agricultural machinery manufacturing enterprises, the technical solution of this system can be integrated into automated grafting equipment to improve the intelligent level and market competitiveness of the equipment;

[0219] 3. For agricultural research institutions, the data and knowledge accumulated by this system can provide a scientific basis for the research and improvement of Camellia oleifera planting technology;

[0220] 4. For agricultural informatization service providers, the technical framework of this system can be extended to other crop grafting applications to expand the business scope.

[0221] In summary, the present invention constructs a complete intelligent diagnosis solution for the defects of Camellia oleifera grafted seedlings by integrating a number of advanced technologies, provides strong support for the modernization and intelligentization of the Camellia oleifera industry, and has broad application prospects and promotion values.

[0222] The above are only the preferred embodiments of the present invention, and do not limit the patent protection scope of the present invention accordingly. Any equivalent structural transformation made under the inventive concept of the present invention by using the content of the specification and drawings of the present invention, or directly / indirectly applied in other related technical fields, shall be similarly included in the patent protection scope of the present invention.

Claims

1. An intelligent diagnosis method for defects of grafted oil-tea camellia seedlings integrating image enhancement, characterized in that Including: Obtain images of oil-tea camellia grafted seedlings and environmental parameters; among them, the environmental parameters include light intensity, temperature, humidity, and plant moisture; Perform adaptive light compensation processing on the images of oil-tea camellia grafted seedlings to obtain enhanced images; including: Based on the environmental parameters, determine light compensation parameters; According to the light compensation parameters, perform region-adaptive enhancement on the images of oil-tea camellia grafted seedlings; Extract and fuse multi-scale features from the enhanced images to obtain feature representations; including: Extract multi-scale spatial features from the enhanced images through a hierarchical feature extraction network; Extract texture features and morphological features through texture-morphology dual-channel analysis; Fuse the multi-scale spatial features, the texture features, and the morphological features to obtain the feature representation; Based on the feature representation and the defect self-attention mechanism, identify defects to obtain defect diagnosis results; Combine the defect diagnosis results with the knowledge graph of oil-tea camellia grafted seedlings for traceability analysis to obtain defect causes and recommended measures; Compare the defect diagnosis results, the defect causes, and the recommended measures with historical data, and output optimization decisions and parameter adjustment plans.

2. The method according to claim 1, wherein The specific process of performing adaptive light compensation processing on the images of oil-tea camellia grafted seedlings includes: Divide the images of oil-tea camellia grafted seedlings into multiple grid regions; Calculate the average brightness value of each grid region; Use a Gaussian mixture model to identify brighteness abnormal regions and construct an illuminance non-uniformity distribution map; Based on the illuminance non-uniformity distribution map, perform illuminance enhancement processing on weak light regions and illuminance suppression processing on overexposed regions respectively.

3. The method according to claim 1, characterized in that, The specific process of extracting and fusing multi-scale features from the enhanced images includes: Adopt a densely connected convolutional network to extract spatial features of different scales and construct a feature pyramid; Use a Gabor filter bank to extract multi-directional and multi-scale texture features; Use a multi-morphological operator sequence to extract morphological features; Design a weighted adaptive fusion algorithm for feature fusion, where the weight coefficients are dynamically adjusted according to the distinctiveness of the features.

4. The method according to claim 1, characterized in that The specific process of identifying defects based on the feature representation and the defect self-attention mechanism includes: The design space self-attention mechanism captures global dependencies, and the formula is According to the attention score matrix A, generate an attention output Z = A·V; Locate the defect regions based on the attention output Z; Adopt a hierarchical cross-entropy loss function for multi-level classification of defect categories.

5. The method according to claim 1, wherein The specific process of combining the defect diagnosis results with the knowledge graph of oil-tea camellia grafted seedlings for traceability analysis includes: Construct an expert knowledge graph including defect types, environmental factors, grafting operations, variety characteristics, and treatment plans; Perform knowledge reasoning using a graph neural network based on the attention mechanism; Use a multi-hop reasoning algorithm to explore the association paths between defects and possible causes; Calculate the conditional probability P(Cause|Defect,Env) based on a Bayesian network; Calculate the causal association strength score, and output the causes and corresponding recommended measures with scores exceeding the threshold.

6. The method according to claim 1, characterized in that, The specific process of comparing the defect diagnosis results, the defect causes, and the recommended measures with historical data includes: Retrieve similar cases from a multi-dimensional historical data warehouse; Construct a multi-objective optimization model: max{w1·survival rate + w2·growth rate + w3·resource efficiency}; Search for the optimal parameter combination based on the particle swarm optimization algorithm; Generate optimization schemes for plant pairing, grafting angle, and binding strength.

7. The method according to claim 1, characterized in that It also includes the step of dynamically adjusting environmental perception parameters: Collect environmental parameter data in real time and perform exponentially weighted moving average processing; Construct a mapping model between the environmental parameter space Θ and the system performance index space Φ; Parameter Adjustment Algorithm Based on Performance Gradient Adjust the processing parameters; Feed back the adjusted processing parameters to the adaptive light compensation and multi-scale feature fusion module.

8. The method according to claim 1, wherein The defect diagnosis results include: Defect types, including normal, graft interface cracking, water shortage and wilting, lesion infection, stem distortion, and malnutrition; Defect location coordinates and area; Defect severity score, ranging from 0 to 1; Defect confidence, with a main category threshold of 0.85 and a sub-category threshold of 0.

75.

9. The method according to claim 1, wherein The hierarchical feature extraction network includes 5 densely connected blocks, each block contains 4 convolutional layers, the growth rate k = 32, forming a feature pyramid structure {64×64×32, 32×32×64, 16×16×128, 8×8×256, 4×4×512}; in the texture-morphology dual-channel analysis, the directions of the Gabor filter bank θ ∈ {0°, 45°, 90°, 135°}, and the scales σ ∈ {1, 2, 4, 8}.

10. An intelligent diagnosis system for defects of grafted oil-tea camellia seedlings with fused image enhancement, which executes the method according to any one of claims 1-9, characterized in that It includes: A data acquisition module for obtaining images of oil-tea camellia grafted seedlings and environmental parameters; An adaptive light compensation module for performing regional adaptive enhancement on the images of oil-tea camellia grafted seedlings to obtain enhanced images; A multi-scale feature fusion module for extracting and fusing multi-scale features from the enhanced images to obtain feature representations; A defect recognition module for recognizing defects based on the feature representations and the defect self-attention mechanism to obtain defect diagnosis results; A defect tracing module for performing tracing analysis by combining the defect diagnosis results with the knowledge graph of oil-tea camellia grafted seedlings to obtain defect causes and recommended measures; A decision optimization module for comparing the defect diagnosis results, the defect causes, and the recommended measures with historical data and outputting optimization decisions and parameter adjustment schemes; An environmental parameter adjustment module for adjusting system processing parameters in real time to adapt to different environmental conditions; Through the collaborative work of the various modules, the system realizes the intelligent diagnosis and decision optimization of defects in oil-tea camellia grafted seedlings.

Images