Seed industry pest early warning method and system based on big data

Abstract

The invention relates to the technical field of disease and insect pest early warning, and discloses a big data-based seed industry disease and insect pest early warning method and system, and the method comprises the steps: obtaining environment sound data and leaf image data; performing scab extraction according to the leaf image data to obtain scab morphological characteristics; performing feature extraction according to the environment sound data to obtain pest sound features; matching according to the scab morphological characteristics and the pest sound characteristics in combination with a pre-stored pest disease database to obtain a pest type and a disease type; performing matching degree calculation according to the pest type and the disease type to obtain a pest and disease matching degree; and carrying out damage assessment according to the pest disease matching degree and a preset matching degree threshold value to obtain a disease assessment result. The method has the following effect that the accuracy of early warning of plant diseases and insect pests in the seed industry can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of pest and disease early warning, and particularly to a method and system for pest and disease early warning of seed industry based on big data. Background Art

[0002] At present, with the development of agricultural technology, methods and systems for pest and disease early warning of seed industry based on big data have gradually become an important means to improve the yield and quality of crops. By integrating multi-source data such as meteorological data, soil information, crop growth status, and historical records of disease occurrence and pest occurrence, the system can respectively achieve accurate prediction and early warning of disease risk and pest risk. Especially for large-scale planting areas, the use of big data technology can not only timely detect the potential outbreak trend of diseases and the occurrence dynamics of pests, but also formulate targeted prevention and control strategies in advance, reduce the usage of chemical pesticides, and protect the ecological environment. In an existing technology, the sound monitoring technology adopted in the pest early warning of the big data-based pest and disease early warning system of the seed industry mainly relies on the unique sound characteristics emitted by pests for identification. The specific process is as follows: First, multiple high-sensitivity microphones are set up in the field to capture various sound signals in the environment; subsequently, these spectrogram data are transmitted to the central processing unit for analysis. With the help of machine learning algorithms, the system can identify the specific audio frequency bands generated when pests fly or move, and match them with the sound patterns of known pests. To enhance the accuracy, the system will also comprehensively judge by combining other data such as weather conditions and time points. Once it is confirmed that there is an abnormal increase trend in pest activities, the system will automatically trigger an alarm and push detailed early warning information to farmers, including recommended measures, etc. This method has greatly improved the efficiency of pest identification.

[0003] For disease early warning, the system mainly analyzes the correlation between environmental data (such as temperature, humidity, rainfall) and crop growth status (such as leaf color, disease spot distribution), and combines historical disease occurrence records to establish a disease risk prediction model. For example, by monitoring the correlation between soil humidity and specific fungal diseases, or using drone image recognition to identify early disease symptoms, the system can early warn of the risk of disease outbreaks and provide non-chemical control suggestions (such as adjusting irrigation strategies or biological control plans).

[0004] Although the above pest warning method has many theoretical advantages, it faces a key challenge in practical applications - the sound characteristics of pests often overlap with the frequencies of environmental noise, making it difficult to accurately separate the two during feature extraction. This is because the sound frequencies produced by many pests are similar to those of wind, rain, or even the sounds of distant mechanical operations in the natural environment, making it very difficult to identify pests solely based on frequency features in a complex acoustic environment. Further, when attempting to remove background noise by increasing filters, the intensity of the sound signals of the target pests will be weakened simultaneously, thereby affecting the final recognition effect. In addition, due to changes in background noise characteristics caused by different geographical regions and seasonal variations, fixed filtering schemes cannot adapt to all situations. This means that existing systems require more intelligent and adaptive methods to distinguish pest sounds from environmental noise, and this limitation restricts the wide application and effectiveness of pest warning systems based on sound monitoring, especially in open farmland environments with high noise levels. In summary, the accuracy of the existing seed industry pest and disease warning methods still needs to be improved during the pest sound extraction process. Summary of the Invention

[0005] The present invention provides a big data-based seed industry pest and disease warning method and system to achieve the goal of improving the accuracy of sound feature extraction in the seed industry pest and disease warning process.

[0006] In a first aspect, to solve the above technical problems, the present invention provides a big data-based seed industry pest and disease warning method, including:

[0007] Obtain environmental sound data and leaf image data;

[0008] Extract lesion spots based on the leaf image data to obtain lesion spot morphological characteristics;

[0009] Extract feature information from the environmental sound data to obtain pest sound characteristics;

[0010] Match the lesion spot morphological characteristics and the pest sound characteristics with a pre-stored pest and disease database to obtain pest types and disease types;

[0011] Calculate the matching degree according to the pest types and the disease types to obtain the pest and disease matching degree;

[0012] Perform a damage assessment based on the pest and disease matching degree and a preset matching degree threshold to obtain a disease assessment result.

[0013] In an optional implementation manner, the extracting lesion spots based on the leaf image data to obtain lesion spot morphological characteristics includes:

[0014] Perform image segmentation on the leaf image data to obtain a lesion spot area image;

[0015] Input the image of the lesion area into a pre-trained lesion extraction model to obtain the morphological characteristics of the lesion;

[0016] Among them, the training process of the lesion extraction model includes:

[0017] Train the lesion extraction model based on a convolutional neural network, with the input layer being historical lesion images and the output layer being a texture shape feature vector;

[0018] When it is detected that the loss function of the model meets the conditions or the number of training times reaches the preset upper limit, the training is completed to obtain the trained model.

[0019] In an alternative implementation, the extracting the pest sound features according to the environmental sound data includes:

[0020] Perform noise reduction filtering on the environmental sound data according to the preset filtering parameters to obtain noise-reduced sound data;

[0021] Perform time-frequency analysis on the noise-reduced sound data to obtain frequency-domain features and time-domain features;

[0022] Perform blind source separation on the noise-reduced sound data where the frequency-domain features overlap with the environmental noise to obtain pure pest sounds;

[0023] Perform time-frequency analysis on the pure pest sounds to obtain pest sound features.

[0024] In an alternative implementation, the matching the pest type and the disease type by combining the lesion morphological characteristics and the pest sound features with a pre-stored pest-disease database includes:

[0025] Calculate the similarity between the lesion morphological characteristics and the disease feature data in the pest-disease database to obtain the disease similarity;

[0026] When the disease similarity is greater than a preset first similarity threshold, it is judged as the disease type corresponding to the disease feature data;

[0027] Calculate the similarity between the pest sound features and the pest feature data in the pest-disease database to obtain the pest similarity;

[0028] When the pest similarity is greater than a preset second similarity threshold, it is judged as the pest type corresponding to the pest feature data.

[0029] In an alternative implementation, the calculating the pest-disease matching degree according to the pest type and the disease type includes:

[0030] Obtain the pest damage vector database;

[0031] Retrieve in the pest damage vector database according to the pest type to obtain the pest damage vector;

[0032] Retrieve in the pest damage vector database according to the disease type to obtain the disease damage vector

[0033] Calculate the cosine correlation degree according to the pest damage vector and the disease damage vector to obtain the pest-disease matching degree.

[0034] In an alternative embodiment, the victimization assessment is performed according to the pest-disease matching degree and a preset matching degree threshold to obtain a disease assessment result, including:

[0035] When the pest-disease matching degree is lower than the preset matching degree threshold, mark the determination results of the pest type and the disease type as unreliable, and re-acquire data;

[0036] When the pest-disease matching degree is higher than the preset matching degree threshold, perform pest and disease distribution analysis according to the disease type, the pest sound characteristics, and the pest type to obtain a pest and disease distribution map;

[0037] Perform density statistics according to the pest and disease distribution map to obtain the pest and disease distribution density;

[0038] Determine the area where the pest and disease distribution density is greater than the preset density threshold as a high-density area;

[0039] Determine the area where the pest and disease distribution density is less than or equal to the preset density threshold as a low-density area;

[0040] Wherein, the disease assessment result includes the disease type, the pest type, and the pest and disease distribution density.

[0041] In an alternative embodiment, after matching according to the disease spot morphological characteristics and the pest sound characteristics with a pre-stored pest and disease database to obtain the pest type and the disease type, it further includes:

[0042] Fuse the disease spot morphological characteristics and the pest sound characteristics to obtain comprehensive feature data;

[0043] Add the comprehensive feature data to the pest and disease database.

[0044] In a second aspect, the present invention provides a big data-based seed industry pest and disease early warning system, including:

[0045] A data acquisition module for acquiring environmental sound data and leaf image data;

[0046] A lesion extraction module for extracting lesions based on the leaf image data to obtain lesion morphological features;

[0047] A pest sound module for extracting features from the environmental sound data to obtain pest sound features;

[0048] A data retrieval module for matching the lesion morphological features and the pest sound features with a pre - stored pest and disease database to obtain the pest type and the disease type;

[0049] A type matching module for calculating the matching degree based on the pest type and the disease type to obtain the pest - disease matching degree;

[0050] An evaluation result module for performing a damage assessment based on the pest - disease matching degree and a preset matching degree threshold to obtain a disease assessment result.

[0051] In a third aspect, the present invention further provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the above - mentioned big - data - based seed industry pest and disease early warning method according to any one of the above is implemented.

[0052] In a fourth aspect, the present invention further provides a computer - readable storage medium, where the computer - readable storage medium includes a stored computer program. When the computer program runs, it controls the device where the computer - readable storage medium is located to execute the above - mentioned big - data - based seed industry pest and disease early warning method according to any one of the above.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] (1) The process of obtaining environmental sound data and leaf image data ensures comprehensive monitoring of the plant health status. By using devices such as high - sensitivity microphones and high - definition cameras to collect the surrounding sound information and detailed images of the leaf surface, rich raw data is provided for subsequent analysis. This process not only improves the quality and integrity of the data but also lays a solid foundation for accurately identifying pests and diseases.

[0055] (2) Extracting lesions based on the leaf image data to obtain lesion morphological features. Using advanced image - processing techniques, such as edge detection, color segmentation, and texture analysis methods, the lesion area is accurately extracted from the complex leaf background, and its shape, size, color, and distribution and other features are quantified. These features provide key bases for identifying different types of diseases and help improve the accuracy of diagnosis.

[0056] (3) Extract features from the environmental sound data to obtain pest sound features. Use an acoustic analysis tool to perform spectral analysis on the sound signals in the environment, identify and separate the unique audio signals emitted by potential pests. By carefully analyzing these sound features, different types of pests and their activity patterns can be distinguished, providing strong support for subsequent pest identification.

[0057] (4) Match the diseased spot morphological features and the pest sound features with a pre-stored pest and disease database to obtain the pest type and disease type. Compare the extracted diseased spot morphological features and pest sound features with the known samples in the database to find the closest matching items. This method can quickly and accurately determine the specific types of pests and diseases, significantly improving the diagnostic efficiency and reliability.

[0058] (5) Calculate the matching degree based on the pest type and the disease type to obtain the pest and disease matching degree. After determining the pest and disease types, further calculate the matching degree of each combination. This step considers various factors, including the cosine similarity of features, constructs an infestation vector, and compares it with the template library for matching, so as to obtain the most corresponding pest and disease types.

[0059] (6) Conduct a damage assessment based on the pest and disease matching degree and a preset matching degree threshold to obtain a disease assessment result. Based on the calculated pest and disease matching degree and referring to the set matching degree threshold, generate pest and disease identification results and the distribution of the damage degree. This assessment method based on scientific analysis not only improves the objectivity of decision-making but also helps farmers take effective control measures in a timely manner, minimizing losses to the greatest extent and ensuring crop yield and quality. In this way, the intelligent monitoring and management of pests and diseases in the agricultural ecosystem are realized, promoting the development of precision agriculture. Description of the Drawings

[0060] Figure 1 is a schematic flowchart of a big data-based seed industry pest and disease early warning method provided by the first embodiment of the present invention;

[0061] Figure 2 is a schematic structural diagram of a big data-based seed industry pest and disease early warning system provided by the second embodiment of the present invention. Detailed Embodiments

[0062] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0063] Reference Figure 1 The first embodiment of the present invention provides a seed industry pest and disease early warning method based on big data, comprising the following steps:

[0064] S11, acquiring environmental sound data and blade image data;

[0065] S12, extracting disease spots according to the leaf image data to obtain disease spot morphological characteristics;

[0066] S13, extracting features according to the environmental sound data to obtain pest sound features;

[0067] S14, matching the lesion morphological features and the pest sound features with a pre-stored pest and disease database to obtain pest types and disease types;

[0068] S15, calculating a matching degree according to the pest type and the disease type to obtain a pest-disease matching degree;

[0069] S16, performing damage assessment according to the pest and disease matching degree and a preset matching degree threshold to obtain a disease assessment result.

[0070] In step S11 , environmental sound data and blade image data are acquired.

[0071] In one embodiment, environmental sound data can be collected by distributed high-sensitivity acoustic sensors (such as MEMS microphone arrays or ultrasonic sensors), whose frequency response range needs to cover 0-100kHz to capture pest activity sound waves. When arranged, the density of 3-5 nodes per hectare is uniformly covered in farmland, and the 30-50cm area below the crop canopy is monitored; leaf image data is obtained by an intelligent camera with multi-spectral imaging capabilities (such as a CMOS sensor with more than 5 million pixels equipped with a near-infrared band). The equipment is installed on a field bracket to form a 45° tilted viewing angle, and cooperates with an automatic rotating pan-tilt to achieve 6-direction scanning of each crop. The two types of data are transmitted to the edge computing gateway in real time through the LoRaWAN Internet of Things protocol, compressed by H.265 encoding, and stored in a local NAS storage system with RAID5 redundancy. At the same time, they are uploaded to the cloud time series database through the 5G network, where the sound data is stored in WAV format in 10-minute segments, and the image data is saved in JPEG2000 format containing EXIF metadata. All data are attached with environmental parameters such as GPS positioning, temperature and humidity sensor readings to form a structured data set.

[0072] In step S12, lesions are extracted based on the leaf image data to obtain morphological features of the lesions.

[0073] Perform image segmentation based on the leaf image data to obtain an image of the lesion area;

[0074] Input the image of the lesion area into a pre-trained lesion extraction model to obtain the morphological features of the lesion;

[0075] Among them, the training process of the lesion extraction model includes:

[0076] Train the lesion extraction model based on a convolutional neural network. The input layer is the historical lesion image, and the output layer is the texture shape feature vector;

[0077] When it is detected that the loss function of the model meets the conditions or the number of training times reaches the preset upper limit, the training is completed to obtain the trained model.

[0078] It should be noted that in the image segmentation implementation process, the specific steps are as follows: First, preprocess the leaf image data, perform smoothing processing on the image using the Gaussian filtering algorithm, and calculate the neighborhood weighted average value pixel by pixel by setting a convolution kernel window of 3×3 or 5×5 to eliminate noise interference. Then use the histogram equalization technique to stretch and adjust the pixel value distribution of each channel of the image, especially enhancing the contrast difference between the lesion area and the healthy leaf in the red channel and the blue channel. Subsequently, convert the RGB color space to the HSV or Lab mode, calculate the joint distribution histogram of the hue component and the saturation component, and find the gray threshold that can best distinguish the lesion from the background. At this time, the system will traverse the gray value range of 0-255 and automatically select the critical value that maximizes the between-class variance of the foreground and background as the segmentation standard to generate a binary mask. After generating the binary mask, perform morphological processing. First, use a 3×3 circular structuring element to perform erosion operation on the image to eliminate isolated noise points, then repair the broken edges through dilation operation, and at the same time use the region filling algorithm to repair the holes inside the lesion. Subsequently, use the edge detection technology to calculate the horizontal gradient and vertical gradient amplitudes of the pixel points, and combine the double-threshold method to determine the strong edges and weak edges, and finally generate a continuous lesion contour line. Implement region growing optimization on the preliminary segmentation result, starting from the seed points on the contour line, and expand the region according to whether the color difference of adjacent pixels is less than the preset threshold (such as the Euclidean distance in the RGB space is less than 15) until all broken regions are connected. Finally, use the connected component labeling algorithm to count each independent lesion area, calculate parameters such as the pixel area, aspect ratio of the circumscribed rectangle, and edge perimeter of each area, and generate an image of the lesion area with coordinate positioning and geometric feature annotation after removing the interference areas with less than 50 pixels.

[0079] It should be noted that the training process of the lesion extraction model is as follows: First, a large number of historical lesion images with annotations are collected as the training set, and the images are standardized, including unifying the resolution to 512×512 pixels, performing Gaussian blur denoising, and implementing data augmentation through random rotation and mirror flipping; Subsequently, a deep convolutional neural network is constructed, whose backbone network adopts a multi-level structure. The first layer extracts edge texture features through a 3×3 convolutional kernel, the second layer compresses the feature dimension through max pooling, and four groups of residual modules are stacked subsequently to capture the complex morphological changes of the lesions, and an attention mechanism module is embedded before the fully connected layer to strengthen the weights of the key regions; During training, a contrast loss function is adopted, and the annotated texture shape feature vector (including 12-dimensional parameters such as area ratio, edge serration degree, and shape irregularity index) is used as the supervision signal. The network parameters are optimized round by round through the backpropagation algorithm. After each iteration, the Euclidean distance between the predicted feature vector and the true label is calculated as the loss value. When the decrease rate of the loss value is less than one-thousandth for 10 consecutive training cycles or the total number of training rounds reaches 200 times, the training is automatically terminated; Finally, the performance of the model is evaluated through the validation set, and the model version with a feature matching error lower than 5% is selected for actual application.

[0080] In step S13, feature extraction is performed on the environmental sound data to obtain pest sound features.

[0081] In one implementation, noise reduction filtering is performed on the environmental sound data according to the preset filtering parameters to obtain noise-reduced sound data;

[0082] Time-frequency analysis is performed on the noise-reduced sound data to obtain frequency-domain features and time-domain features;

[0083] Blind source separation is performed on the noise-reduced sound data where the frequency-domain features overlap with the environmental noise to obtain pure pest sounds;

[0084] Time-frequency analysis is performed on the pure pest sounds to obtain pest sound features.

[0085] It should be noted that the specific implementation of the extraction process of pest sound characteristics is as follows: First, based on preset filtering parameters (such as the passband frequency range of 200 Hz - 8 kHz and the stopband attenuation of 40 dB), adaptive noise reduction filtering is performed on the original environmental sound data, and a Butterworth bandpass filter bank is used to suppress low-frequency mechanical vibration noise and high-frequency electromagnetic interference; then, the short-time Fourier transform (STFT) is performed on the noise-reduced sound signal, with a 25-ms Hamming window and a 10-ms step size set to extract 24-dimensional time-frequency characteristics including Mel-frequency cepstral coefficients (MFCC), spectral centroid, and zero-crossing rate, and at the same time, the root mean square energy, pulse duration, and repetition frequency in the time domain are calculated; for the mixed signal where the frequency-domain characteristics overlap with environmental noise (such as wind noise and rain noise), an improved independent component analysis (ICA) algorithm is used for blind source separation, and by maximizing the non-Gaussianity measure between signals, the pure acoustic signal of the target pest is separated from the multi-channel recording; finally, a refined time-frequency analysis is performed on the separated signal, and the continuous wavelet transform (CWT) is used to extract the pest sound characteristics of pest gnawing or wing flapping in the range of 1 - 12 kHz with the Morlet wavelet basis function, including key bioacoustic parameters such as the main frequency band energy distribution (such as Lepidoptera pests concentrated in 3 - 5 kHz), the time interval of the pulse sequence (about 50 - 200 ms for Coleoptera), and the harmonic attenuation slope.

[0086] In one implementation, in the blind source separation stage, the system first performs multi-scale time-frequency analysis on the noise-reduced mixed audio. The continuous audio is segmented into 50-ms segments using a sliding window mechanism and a Hamming window is superimposed, and a three-dimensional spectrogram (frequency - time - energy) is generated through the short-time Fourier transform. For the characteristic of the 3.5-kHz main frequency pulse group unique to Cnaphalocrocis medinalis, the system constructs a dynamic mask template: in the frequency spectrum dimension, morphological dilation operations are performed on the harmonic clusters with periodic mutations in the 2 - 5 kHz frequency band (such as 5 consecutive peaks appearing at an interval of 0.2 s) to enhance the target signal; in the time dimension, a time-domain energy baseline is established using the diurnal activity pattern of pests (such as the nocturnal calling intensity of Chilo suppressalis increasing by 40%). Subsequently, the non-negative matrix factorization technique is used to decompose the mixed spectrum into a basis matrix (representing the characteristics of insect calls) and a coefficient matrix (representing environmental noise), and through iterative optimization, the cosine similarity between the decomposed insect call basis and the preset voiceprint feature library (such as the standard calling patterns of 20 Lepidoptera insects) reaches more than 85%. For the remaining broadband noise (such as wind noise), the system enables an adaptive spectral subtraction algorithm: according to the median value of the noise floor energy in adjacent 10 time-frequency windows, the attenuation coefficient of each frequency point is dynamically adjusted, and finally, the pure insect call signal is reconstructed through the inverse Fourier transform while retaining key biological characteristics (such as the 12-Hz modulation ripple of the mating signal).

[0087] In step S14, based on the morphological characteristics of the lesion and the pest sound characteristics, a match is made with a pre-stored pest and disease database to obtain the pest type and disease type.

[0088] In one implementation, a similarity calculation is performed based on the morphological characteristics of the lesion and the disease characteristic data in the pest and disease database to obtain a disease similarity; when the disease similarity is greater than a preset first similarity threshold, it is determined as the disease type corresponding to the disease characteristic data; a similarity calculation is performed based on the pest sound characteristics and the pest characteristic data in the pest and disease database to obtain a pest similarity; when the pest similarity is greater than a preset second similarity threshold, it is determined as the pest type corresponding to the pest characteristic data.

[0089] It should be noted that when the disease similarity or pest similarity does not reach the preset threshold, the system will adopt different strategies for subsequent processing: for diseases, if the similarity is lower than the threshold, it indicates that the current lesion morphology has no significant match with the known disease characteristics in the database. The system will mark it as "unknown disease type" and trigger a secondary verification process, such as supplementing the collection of multi-angle images of the lesion or comprehensively analyzing in combination with environmental data (such as humidity, temperature), and at the same time push a prompt message to the farmer, suggesting manual re-inspection or providing samples for inspection; for pests, if the similarity does not reach the threshold, the system determines it as "unable to identify the pest type", starts the re-collection of sound signals or extends the monitoring duration to obtain more complete data, and synchronously combines other sensor information (such as the degree of crop damage, environmental vibration frequency) for cross-verification. Finally, if it still cannot be determined, a warning prompt is sent to the farmer, suggesting strengthening field inspections or adopting broad-spectrum preventive measures, and at the same time storing the unrecognized data in the database for subsequent model optimization reference.

[0090] It should be noted that the identification and matching process of pests and diseases is realized based on a multi-modal feature fusion mechanism: First, the pre-stored pest and disease database is called. This database contains the lesion morphological parameters of more than 300 typical diseases (such as edge serration density, lesion expansion directionality, color gradient pattern) and the acoustic feature parameters of more than 200 pests (such as vibration fundamental frequency, pulse interval regularity, harmonic attenuation characteristics); for disease identification, the system compares the lesion morphological features collected in real time with each disease template in the database item by item, and calculates the similarity through the weighted comprehensive scoring method (for example, the weight of the lesion area ratio is 30%, the texture complexity is 25%, and the edge sharpness is 45%). When the comprehensive similarity of a certain disease exceeds the 85% threshold and is 5 percentage points higher than other candidate diseases, it is determined as this disease type; in the pest identification link, the dynamic time warping algorithm is used to perform time series alignment processing on the collected acoustic features and the standard acoustic features in the database, and the two core indicators of pulse interval stability (weight 40%) and spectral energy distribution (weight 35%) are mainly matched. When the matching degree reaches 90%, the determination is triggered, that is, it is determined as this pest type; when there are conflicts in the two types of identification results (such as the disease pointing to fungal infection and the pest pointing to chewing mouthpart pests), the system will start the multi-modal decision-making module and preferentially adopt the determination result with a confidence difference exceeding 15%. If the difference is within the range of 10%, a composite pest and disease warning will be output; a double verification mechanism is set for the entire matching process. When the single feature matching degree exceeds the threshold but the comprehensive confidence is less than 75%, the secondary feature extraction process will be automatically triggered to exclude misjudgments caused by environmental interference.

[0091] It is worth noting that the dynamic time warping algorithm (DTW) performs time series alignment processing on the collected voiceprint features and the standard acoustic features in the database through the following steps: First, the voiceprint signal is converted into time series data, which contains two core features: pulse interval (such as the interval time between the pests flapping their wings or gnawing) and spectrum energy distribution (the energy proportion of each frequency component). The DTW algorithm constructs a two-dimensional distance matrix, calculates the local distance (such as Euclidean distance) at each time point, and uses a dynamic programming strategy to find the global optimal alignment path, which allows the time series to be elastically scaled on the time axis to adapt to the rhythm or frequency offset caused by individual differences or environmental interference of different pests. For example, if the pulse interval of the standard voiceprint of a pest is relatively stable (such as the regularity of the flapping frequency), and the actual collected voiceprint has a slightly longer interval due to the influence of wind speed, DTW can align the pulse patterns of the two through path adjustment. The algorithm also gives a weight of 40% to the pulse interval stability and 35% to the spectrum energy distribution, and comprehensively calculates the matching score of the two. When the total match (such as the weighted similarity percentage) reaches 90%, the system determines that the two voiceprints are highly matched, thereby confirming the type of pest. This process solves the problem of time series misalignment through nonlinear alignment, allowing the algorithm to more accurately identify the similarity of core features rather than simply relying on strict time synchronization.

[0092] In a specific embodiment, a rice field system achieves accurate recognition through a multimodal feature fusion mechanism: when a leaf is detected with a serrated brown spot (spot diameter 1.5 cm, 28% of the leaf area, and edge serration density 8 teeth / cm 2 The system performed weighted calculations based on the proportion of diseased area (30% weight), texture complexity (25% weight) and edge sharpness (45% weight), and obtained a similarity of 86.7% to rice blast, which was significantly higher than other candidate diseases. At the same time, the collected 3.2kHz soundprint signal was aligned by the dynamic time warping algorithm and matched to a pulse interval stable at 120±5ms (40% weight), a 3.2kHz frequency band energy proportion of 65% (35% weight) and a harmonic attenuation slope of 0.8dB / oct (25% weight), and was finally determined to be a rice stem borer pest with a 90% match. As the confidence difference between the two is only 3.3% and the typical correlation between the pest type (chewing mouthparts) and the disease (rice blast) is weak, the system activates the multimodal decision-making module and issues a composite warning (the main alert is rice blast, and the secondary alert is that the activity of the Chilo suppressalis will aggravate the spread of fungi). In response to the soundprint interference caused by 95% ambient humidity, the system conducted double verification through retesting the sound signal in the early morning and re-inspection of the mold spores on the lesions, and finally maintained a judgment conclusion with a comprehensive confidence of 78%.

[0093] In step S15, a matching degree is calculated based on the pest type and the disease type to obtain a pest-disease matching degree.

[0094] In one implementation, an infringement vector database is obtained;

[0095] Retrieve in the infringement vector database according to the pest type to obtain a pest infringement vector;

[0096] Retrieve in the infringement vector database according to the disease type to obtain a disease infringement vector

[0097] Calculate the cosine correlation degree according to the pest infringement vector and the disease infringement vector to obtain the pest-disease matching degree.

[0098] It should be noted that first, an infringement vector database containing various biological infringement characteristics needs to be constructed in advance. Each vector in this database consists of multiple dimensions, such as the parts of the crop being damaged (such as leaves, roots), the occurrence season, the transmission route (such as insect vectors, soil), typical symptoms (such as spots, wilt), and other key attributes. When the system obtains the pest type and disease type input by the user, it will respectively retrieve the corresponding pest infringement vector and disease infringement vector in the infringement vector database. These two vectors respectively quantify the characteristic manifestations of pests and diseases in various attributes. Subsequently, the system calculates the similarity of the two vectors through the cosine correlation algorithm. This algorithm judges the consistency of their directions by analyzing the numerical distribution trends of the two vectors in each dimension: if the included angle between the directions of the two vectors is smaller (i.e., the cosine value is closer to 1), it means that the matching degree between the pest and the disease in terms of damage characteristics is higher, and vice versa. This calculation process finally outputs the pest-disease matching degree value, which is used to evaluate the potential correlation between the two.

[0099] It should be noted that the cosine correlation formula is as follows:

[0100]

[0101] Among them, cos sim (A,B) represents the cosine correlation degree of vectors A and B, A·B represents the dot product of vectors A and B, ∥A∥ represents the norm of vector A, and ∥B∥ represents the norm of vector B.

[0102] In a specific embodiment, taking aphids and cucumber mosaic virus as examples, the invasion vectors can be specifically manifested as follows: The invasion vector of aphids includes four dimensions: the damaged parts (leaves, tender shoots), the high-incidence seasons (spring and summer), the transmission routes (insect vectors), and the typical symptoms (leaf curling), and different numerical weights are assigned to each dimension; the invasion vector of cucumber mosaic virus includes four dimensions: the damaged parts (leaves, flowers), the high-incidence seasons (from late spring to summer), the transmission routes (insect vectors), and the typical symptoms (yellow spots, deformities). When calculating the matching degree, the system will compare each dimension of the two vectors - for example, their transmission routes are exactly the same, the seasons highly overlap, the damaged parts partially coincide, and the symptoms are related. By analyzing the degree of coordination of the numerical changes in each dimension (such as both reaching peaks in the transmission route dimension and having similar curves in the season dimension), the overall direction consistency is judged, and finally a relatively high matching degree value (such as 0.85) is obtained, reflecting a high probability of aphids transmitting the virus.

[0103] In step S16, a damage assessment is performed according to the pest-disease matching degree and a preset matching degree threshold to obtain a disease assessment result.

[0104] In one implementation manner, when the pest-disease matching degree is lower than the preset matching degree threshold, mark the determination results of the pest type and the disease type as unreliable, and re-obtain data;

[0105] When the pest-disease matching degree is higher than the preset matching degree threshold, perform a pest and disease distribution analysis according to the disease type, the pest sound characteristics, and the pest type to obtain a pest and disease distribution map;

[0106] Perform density statistics according to the pest and disease distribution map to obtain the pest and disease distribution density;

[0107] Determine the areas where the pest and disease distribution density is greater than the preset density threshold as high-density areas;

[0108] Determine the areas where the pest and disease distribution density is less than or equal to the preset density threshold as low-density areas;

[0109] Among them, the disease assessment result includes the disease type, the pest type, and the pest and disease distribution density.

[0110] During the process of pest and disease distribution analysis, the system first spatially associates the identified disease types (such as fungal diseases, viral diseases) and pest types (such as aphids, locusts) with the corresponding geographical coordinate information (such as the location of sensors or monitoring points). Specifically, the system binds the disease type, pest type, and their acoustic characteristics (such as pest activity frequency, voiceprint intensity) of each monitoring point to the geographical coordinates, forming a multi-dimensional spatial data set. Subsequently, through spatial clustering algorithms (such as K-means or DBSCAN) or rasterization methods, the farmland area is divided into fixed or dynamic grid cells, and the comprehensive occurrence density of diseases and pests in each cell is statistically calculated. For example, if multiple diseases or pests are detected simultaneously in a grid cell, the system will calculate the comprehensive density value by weighting according to their damage degree; if there is only a single disease or pest, its occurrence frequency will be directly counted. Finally, the system generates a heat map (i.e., pest and disease distribution map) superimposed with disease types, pest types, and density values, and divides the high-density area (requiring urgent treatment) and low-density area (routine monitoring) based on a preset density threshold (such as the average damage index per square meter exceeding 3.0). This process combines spatial statistics and machine learning models, taking into account both the type differences of pests and diseases and quantifying their spatial aggregation degree, providing a visual basis for precise prevention and control. For example, if a high-density area has both sheath blight of rice (disease) and planthoppers (pests) at the same time, the system will mark this area as a "double high-risk area" and recommend the combined use of biological pesticides and physical control measures.

[0111] In one implementation, when the system determines that the pest-disease matching degree is higher than the preset threshold, it first accurately locates the spatial coordinates of the sound source based on the pest calling characteristics (such as the intermittent high-frequency sound waves unique to the oriental migratory locust, the continuous low-frequency vibrations of the cotton bollworm at night) and their activity patterns (such as the collective calling of the rice leaf folder at dusk) through the Internet of Things acoustic sensor array deployed in the farmland, and generates a heat map that reflects the density difference with color gradients - dark red indicates the core pest aggregation area where more than 100 sound signal triggers are detected per square meter, and light yellow indicates the edge area with less than 20 triggers. Then, the farmland is divided into 10-meter × 10-meter grid cells, and the total number of sound signal triggers per hour in each grid is statistically calculated, and the activity intensity is calibrated in combination with the pest body size coefficient (such as the single call coverage range of the larger locust reaches 3 square meters). Finally, the standardized density index is calculated. For example, if a grid records 800 triggers and the calibration coefficient is 0.8, the density index is 640. When it significantly exceeds the preset threshold of 500, the system automatically marks this area as a high-density red warning area; while the grids with an index between 200 - 500 are marked as yellow monitoring areas, and those with an index lower than 200 are marked as green safe areas.

[0112] It should be noted that when calculating the density index of pests and diseases, the system first counts the total number of acoustic signal triggers per hour in each grid cell (such as the number of times the sound wave generated by pest chirping or activity triggers the sensor), which reflects the real-time activity intensity of pests in this area. However, the difference in body size of different pests will lead to different sound coverage ranges for individual individuals (for example, a single chirp of a large locust can cover 3 square meters, while the sound wave coverage range of small pests such as aphids is only 0.1 square meters). Therefore, it is necessary to calibrate the acoustic signal data through the "body size coefficient": specifically, the system divides the total number of acoustic signal triggers per hour in the grid by the single sound wave coverage area of this pest (i.e., the body size coefficient) to estimate the standardized density index of pests in this grid. For example, if 100 locust acoustic signal triggers are detected per hour in a certain grid, since its coverage area is 3 square meters, the standardized density index is 33.3 (unit: pests per square meter), while for aphids with the same trigger amount, it is 1000 (unit: pests per square meter). This process unifies the activity intensity of different pests into a comparable "standardized density index" by eliminating the influence of body size differences on the sound signal coverage range, thus more accurately reflecting the actual pest density distribution.

[0113] In summary, the present invention discloses a method for early warning of pests and diseases in the seed industry based on big data, aiming to improve the accuracy of early warning of crop pests and diseases by integrating multiple data sources and advanced analysis techniques. This method first involves obtaining environmental sound data and leaf image data, collecting the surrounding sound information and detailed images of the leaf surface through devices such as high-sensitivity microphones and high-definition cameras, providing rich raw data for subsequent analysis. Then, according to the obtained leaf image data, disease spots are extracted. Advanced image processing techniques such as edge detection, color segmentation, and texture analysis are used to accurately extract the disease spot area from the complex leaf background and quantify its characteristics such as shape, size, color, and distribution. These characteristics provide key bases for identifying different types of diseases. At the same time, after the environmental sound data undergoes feature extraction, spectral analysis is performed on the sound signals in the environment using acoustic analysis tools to identify and separate the unique audio signals emitted by potential pests. By carefully analyzing these sound characteristics, different types of pests and their activity patterns can be distinguished, providing strong support for subsequent pest identification.

[0114] Furthermore, the method proposed by the present invention matches the extracted morphological characteristics of disease spots and the sound characteristics of pests with a pre-stored pest and disease database, and uses machine learning or pattern recognition algorithms to find the closest matching items to quickly and accurately determine the specific types of pests and diseases. In this process, not only the morphological characteristics of the disease spots are considered, but also the specific sound patterns produced by the pests. This dual verification mechanism greatly improves the accuracy of diagnosis. After determining the types of pests and diseases, the system further calculates the matching degree of each combination. This process takes into account various factors, including feature similarity, occurrence probability, etc., so as to assign a quantitative score to each matching result. This scoring mechanism helps to evaluate the credibility of the diagnostic results and guide subsequent damage assessment. Based on the calculated pest and disease matching degree and referring to the set matching degree threshold, the system can automatically judge the degree of crop damage and generate a detailed disease assessment report.

[0115] Referring to Figure 2 , the second embodiment of the present invention provides a big data-based seed industry pest and disease early warning system, including:

[0116] A data acquisition module for acquiring environmental sound data and leaf image data;

[0117] A disease spot extraction module for extracting disease spots according to the leaf image data to obtain morphological characteristics of disease spots;

[0118] A pest sound module for extracting features according to the environmental sound data to obtain pest sound characteristics;

[0119] A data retrieval module for matching according to the morphological characteristics of the disease spots and the pest sound characteristics in combination with a pre-stored pest and disease database to obtain pest types and disease types;

[0120] A type matching module for calculating the matching degree according to the pest type and the disease type to obtain the pest and disease matching degree;

[0121] An evaluation result module for performing damage assessment according to the pest and disease matching degree and a preset matching degree threshold to obtain a disease assessment result.

[0122] It should be noted that the big data-based seed industry pest and disease early warning system provided by the embodiment of the present invention is used to execute all the process steps of the big data-based seed industry pest and disease early warning method in the above embodiment. The working principles and beneficial effects of the two correspond one by one, so they will not be elaborated here.

[0123] An embodiment of the present invention also provides an electronic device. The electronic device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a data acquisition program. When the processor executes the computer program, the steps in the above-mentioned embodiments of each big data-based seed industry pest warning method are implemented, such as Figure 1 the step S11 shown. Alternatively, when the processor executes the computer program, the functions of each module / unit in the above-mentioned device embodiments are implemented, such as the data acquisition module.

[0124] Exemplarily, the computer program may be divided into one or more modules / units. The one or more modules / units are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program in the electronic device.

[0125] The electronic device may be a computing device such as a desktop computer, a notebook, a palm computer, and a smart tablet. The electronic device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above components are only examples of the electronic device and do not constitute a limitation on the electronic device. It may include more or fewer components than the above, or combine certain components, or different components. For example, the electronic device may further include input / output devices, network access devices, a bus, etc.

[0126] The so-called processor may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The processor is the control center of the electronic device and connects various parts of the entire electronic device through various interfaces and lines.

[0127] The memory can be used to store the computer programs and / or modules. By running or executing the computer programs and / or modules stored in the memory, and invoking the data stored in the memory, the processor can implement various functions of the electronic device. The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store the operating system, application programs required for at least one function (such as the sound playback function, the image playback function, etc.); the data storage area can store the data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory can include high-speed random access memory, and can also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0128] Among them, if the modules / units integrated in the electronic device are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the above method embodiments of the present invention, it can also be completed by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disc, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0129] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0130] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. In particular, for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for early warning of plant diseases and insect pests in the seed industry based on big data, characterized in that, Including: Obtain environmental sound data and leaf image data; Extract lesion spots according to the leaf image data to obtain lesion spot morphological characteristics; Extract features from the environmental sound data to obtain pest sound characteristics; Match according to the lesion spot morphological characteristics and the pest sound characteristics with a pre-stored pest and disease database to obtain pest types and disease types; Calculate the matching degree according to the pest type and the disease type to obtain the pest and disease matching degree; Perform damage assessment according to the pest and disease matching degree and a preset matching degree threshold to obtain a disease assessment result.

2. The method for early warning of plant diseases and insect pests in the seed industry based on big data according to claim 1, characterized in that, The extracting lesion spots according to the leaf image data to obtain lesion spot morphological characteristics includes: Perform image segmentation on the leaf image data to obtain a lesion spot area image; Input the lesion spot area image into a pre-trained lesion spot extraction model to obtain lesion spot morphological characteristics; Among them, the training process of the lesion spot extraction model includes: Train the lesion spot extraction model based on a convolutional neural network, with the input layer being historical lesion spot images and the output layer being a texture shape feature vector; When it is detected that the loss function of the model meets the conditions or the number of training times reaches a preset upper limit, the training is completed to obtain a trained model.

3. The seed industry pest and disease warning method based on big data according to claim 1, wherein The extracting features from the environmental sound data to obtain pest sound characteristics includes: Perform noise reduction filtering on the environmental sound data according to preset filtering parameters to obtain noise-reduced sound data; Perform frequency domain analysis on the noise-reduced sound data to obtain frequency domain characteristics; Perform blind source separation on the noise-reduced sound data where the frequency domain characteristics overlap with environmental noise to obtain pure pest sounds; Perform time-frequency analysis on the pure pest sounds to obtain pest sound characteristics.

4. The method for early warning of plant diseases and insect pests in the seed industry based on big data according to claim 1, wherein, The matching according to the lesion spot morphological characteristics and the pest sound characteristics with a pre-stored pest and disease database to obtain pest types and disease types includes: Calculate the similarity between the lesion spot morphological characteristics and the disease characteristic data in the pest and disease database to obtain a disease similarity; When the disease similarity is greater than a preset first similarity threshold, it is judged as the disease type corresponding to the disease characteristic data; Calculate the similarity between the pest sound characteristics and the pest characteristic data in the pest and disease database to obtain a pest similarity; When the pest similarity is greater than a preset second similarity threshold, it is judged as the pest type corresponding to the pest characteristic data.

5. The method for early warning of seed industry pests and diseases based on big data according to claim 1, characterized in that, The calculating the matching degree according to the pest type and the disease type to obtain the pest and disease matching degree includes: Obtain an infringement vector database; Retrieve according to the pest type in the infringement vector database to obtain a pest infringement vector; Retrieve according to the disease type in the infringement vector database to obtain a disease infringement vector; Calculate the cosine correlation degree according to the pest infringement vector and the disease infringement vector to obtain the pest and disease matching degree.

6. The method for early warning of plant diseases and insect pests in the seed industry based on big data according to claim 1, wherein The performing damage assessment according to the pest and disease matching degree and a preset matching degree threshold to obtain a disease assessment result includes: When the pest-disease matching degree is lower than the preset matching degree threshold, mark that the determination results of the pest type and the disease type are unreliable, and re-acquire data; When the pest-disease matching degree is higher than the preset matching degree threshold, perform pest and disease distribution analysis based on the disease type, the pest sound characteristics, and the pest type to obtain a pest and disease distribution map; Perform density statistics based on the pest and disease distribution map to obtain the pest and disease distribution density; Determine the area where the pest and disease distribution density is greater than the preset density threshold as a high-density area; Determine the area where the pest and disease distribution density is less than or equal to the preset density threshold as a low-density area; Wherein, the disease assessment result includes the disease type, the pest type, and the pest and disease distribution density.

7. The method for early warning of seed industry pests and diseases based on big data according to claim 1, wherein After matching the disease spot morphological characteristics and the pest sound characteristics with the pre-stored pest-disease database to obtain the pest type and the disease type, it further includes: Fuse the disease spot morphological characteristics and the pest sound characteristics to obtain comprehensive feature data; Add the comprehensive feature data to the pest-disease database.

8. A seed industry pest and disease early warning system based on big data, characterized in that, It includes: A data acquisition module for acquiring environmental sound data and leaf image data; A disease spot extraction module for extracting disease spots based on the leaf image data to obtain disease spot morphological characteristics; A pest sound module for extracting characteristics from the environmental sound data to obtain pest sound characteristics; A data retrieval module for matching the disease spot morphological characteristics and the pest sound characteristics with the pre-stored pest-disease database to obtain the pest type and the disease type; A type matching module for calculating the matching degree based on the pest type and the disease type to obtain the pest-disease matching degree; An evaluation result module for performing damage assessment based on the pest-disease matching degree and the preset matching degree threshold to obtain a disease assessment result.

9. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the big data-based seed industry pest and disease early warning method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the big data-based seed industry pest and disease early warning method according to any one of claims 1 to 7.

Images