Paddy rice stem borer peak period detection method based on intelligent forecasting lamp data

Abstract

The invention discloses a rice stem borer peak period detection method based on intelligent forecasting lamp data. The method comprises the following steps: step 1, acquiring and preprocessing insect pest monitoring time sequence data; 2, identifying and repairing abnormal trapping data based on the working condition characteristics of the insect trapping device; step 3, constructing a multi-peak Gaussian model; step 4, model adaptive optimal parameter selection and peak period detection; step 5, peak value merging and ecological generation division; and step 6, peak period information extraction and prevention and control window output. According to the method, the overall trend and peak value characteristics of the time sequence curve are utilized, local interference can be effectively reduced, and the accuracy and stability of moth peak recognition are improved; and multiple moth peak characteristics of multiple sites and multiple years in the region can be synchronously extracted, measurement standards do not need to be independently set for different scenes, and the method can be applied to large-range chilo suppressalis population dynamic monitoring.

Description

Technical Field

[0001] This invention relates to the field of agricultural pest and disease monitoring and early warning technology, specifically a method for detecting peak rice stem borer activity based on data from intelligent monitoring lamps. Background Technology

[0002] Monitoring rice stem borers is crucial for rice cultivation. Taking the rice stem borer (Chilodon grandiflorus) as an example, there are currently two main types of monitoring and data processing technologies. The manual field survey method involves plant protection personnel sampling at fixed points in the fields every 3-7 days, recording indicators such as the number of egg masses and pupation rate to estimate the peak adult emergence period. While this method can reflect the development process of the pest in a local field to some extent, it relies excessively on manual labor, is time-consuming and labor-intensive, and the data is highly subjective. It can only cover a small number of representative fields and cannot achieve continuous dynamic monitoring at a regional scale, lacking timeliness and adaptability to large-scale applications.

[0003] To address the challenge of manual surveys in meeting the needs of continuous, dynamic pest monitoring at the regional scale, intelligent pest monitoring lamps, as an automated monitoring method, are widely used for monitoring the occurrence of important pests such as the rice stem borer. This technology acquires daily time-series data on the number of insects attracted using automatic or semi-automatic monitoring lamps, providing a data foundation for pest trend analysis, threshold warnings, and research on their relationship with meteorological factors. However, in actual long-term field operation, the data collected by the monitoring lamps is easily interfered with by various abiotic factors, resulting in high noise levels in the original time-series signal of the attracted insects, making it difficult to directly use for accurate determination of the peak occurrence period and control window for the stem borer. On the one hand, equipment malfunctions such as power supply abnormalities and communication interruptions can cause missing or abnormally low insect attraction data; on the other hand, extreme weather conditions such as heavy rainfall and strong winds can significantly reduce the rate of adult insects reaching the lamps, causing the attracted insects to deviate from the actual field pest population level in the short term, resulting in a large number of low-value fluctuations inconsistent with biological processes, further increasing the difficulty of extracting effective pest information.

[0004] To address the aforementioned issues, existing technologies primarily rely on manual experience analysis or simple time-series smoothing. One type of method judges pest development trends by manually dividing generation intervals or peak times. While this can identify some concentrated peaks, it is highly subjective, lacks standardized criteria, and is difficult to adapt to large-scale applications across multiple sites and years. It also cannot reliably identify the duration of peaks or preceding upward signals. Another type of method uses fixed-window moving averages or simple filtering algorithms to smooth daily data. While suppressing random fluctuations, this often weakens peak amplitudes or even causes peak shifts, hindering the utilization of high temporal resolution data and lacking ecological constraints corresponding to pest occurrence patterns. Furthermore, this type of method is essentially pure data smoothing, lacking ecological constraints corresponding to the pest occurrence process, making it difficult to interpret the biological significance of peaks.

[0005] Because the above methods do not systematically correct for non-biological low values ​​and weather-induced anomalies, nor do they finely characterize the multi-peak structure, existing technologies generally have difficulty stably identifying the start and end range, duration, and multi-generation peak structure of rice stem borer peaks. This restricts the application effect of intelligent monitoring lamp data in regional-scale precise monitoring and control decision-making, and makes it difficult to use a large-scale automated monitoring lamp network to achieve accurate detection of rice stem borer peaks and early warning of control windows. Summary of the Invention

[0006] The purpose of this invention is to provide a method for detecting the peak period of rice stem borers based on data from intelligent monitoring lamps, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for detecting peak rice stem borer activity based on data from intelligent monitoring lamps includes: Step 1: Acquire and preprocess the time-series data of insect attraction quantity; Step 2: Identify and repair abnormal trapping data based on the operating characteristics of the insect-attracting device; Step 3: Construct a multi-peak Gaussian model; Step 4: Perform adaptive optimal parameter selection and peak detection for the multi-peak Gaussian model; Step 5: Perform peak merging and ecological generation classification; Step 6: Extract peak period information and output prevention and control information.

[0008] Further, step 1 includes: Based on the distribution data of regional pest monitoring stations, raw data on the number of insects attracted by all pest monitoring devices in the target area were collected and integrated to obtain time-series data on the number of insects attracted by each device.

[0009] Further, step 2 includes: An adaptive threshold method was used to identify abnormally low values ​​in the time series data of insect attraction, and to distinguish between abnormal operating conditions of the monitoring lamps and biological fluctuations of the borers. For all identified abnormally low value intervals, cubic spline interpolation and boundary constraint repair were used to generate the time series curve of insect attraction after abnormally low value interpolation.

[0010] Furthermore, the adaptive threshold method in step 2 is as follows: The adaptive threshold is dynamically calculated based on the median, 25th percentile, and annual maximum value of growing season data, using the following formula: Adaptive Threshold = max(α·global_max, β·global_median, global_q25, T) min ); In the formula, `max` represents the maximum value among the four indicators described later; `global_max` is the global maximum value, `global_median` is the median of the growing season, `global_q25` is the 25th percentile of the occurrence season, and `T`... min The minimum fixed threshold is α, the global maximum adjustment coefficient is β, and the median adjustment coefficient is β. To preserve the original data trend and avoid introducing spurious peaks during anomaly repair, the 90th quantiles of the data from 30 days before and after the anomaly interval are connected, and then Loess smoothing is used to form the upper envelope. For any time t within the anomaly interval, the interpolated repair value yinterp(t) must satisfy: yinterp(t)≤ γ·yupper(t) Where yinterp(t) is the repair value obtained by cubic spline interpolation at time t; yupper(t) is the value of the upper envelope at time t; and γ is the upper envelope relaxation coefficient.

[0011] Further, step 3 includes: The time-series curve of insect attraction after repair was smoothed by SG filtering, and potential peak values ​​were initially detected based on the smoothed curve. A multi-peak Gaussian model was constructed, and the model parameters were initialized based on the initial peak values. The model parameters were optimized by curve fitting to generate a fitting curve that closely matches the measured data. The formula for the Gaussian multi-peak model is as follows: , Where t is the time variable, representing the cumulative number of days within the monitoring period; y(t) is the daily insect trapping amount obtained by fitting the Gaussian multi-peak model on day t; n is the number of peaks, p i Let exp represent the intensity of the i-th peak, and let t be the natural exponential function. 0i For peak time, w i The peak width is defined as follows: The multi-peak Gaussian model is composed of multiple Gaussian components superimposed on each other, with each component corresponding to a pest occurrence peak. The model parameters are optimized by fitting using the least squares method to ensure that the mean square error between the fitted curve and the measured data is minimized.

[0012] Further, step 4 includes: For the time series curve of insect attraction, a candidate set of SG filter parameters is set; for each parameter combination, the model building process in step 3 is executed to calculate the determination coefficient of Gaussian fitting, Bayesian information criterion and peak position stability index for each combination, and the parameter combination with the best overall performance is selected.

[0013] Furthermore, in step 4, the coefficient of determination R 2 The Bayesian Information Criterion (BIC) and the peak position stability index (peakvar) are calculated as follows: Coefficient of determination R 2 : , , , Where N is the number of samples, y obs For the original observation value, y fit These are Gaussian fitted values. RSS is the arithmetic mean of all original observations; RSS is the residual sum of squares, which is the sum of squares of the differences between the original observations and the Gaussian fitted values; SST is the total sum of squares, which is the sum of squares of the differences between the original observations and the mean of the original observations. Bayesian Information Criterion (BIC): , Peak position stability index peakvar: , Where n is the number of peaks; k is the total number of independent parameters of the model, and each Gaussian fitting peak contains 3 Gaussian parameters, k=3n; , These represent the positions of the i-th peak and the (i-1)-th peak after sorting by time, respectively. Indicator normalization score: The coefficient of determination R 2 After normalization using the Bayesian Information Criterion (BIC) and the peak position stability index (peakvar), a final score is synthesized according to weights, and the window with the highest score is selected as the optimal parameter. , Among them, w1, w2, and w3 are the weight coefficients of the Bayesian Information Criterion (BIC), the Coefficient of Determination (R²), and the Peak Position Stability Index (peakvar), respectively.

[0014] Further, step 5 includes: Based on the fitting results corresponding to the optimal parameters, each local peak is extracted, and the intensity, peak time, peak width, and corresponding 95% time confidence interval of each peak are calculated. By performing overlap checks on the time confidence intervals of different peaks, redundant peaks caused by noise or overfitting are filtered out, and statistically significant effective peaks are retained. For the selected effective peaks, generation matching is performed in conjunction with the occurrence pattern of rice stem borer.

[0015] Furthermore, the peak selection logic in step 5 is based on a 95% confidence interval overlap check. If the time confidence intervals of two peaks overlap, the peak with higher intensity is retained. The generation label assignment adds artificial ecological constraints, and the final label is determined by the matching degree between the peak time and the generation range division.

[0016] Further, step 6 includes: When at least one valid peak is detected within the monitoring period, it is determined that the pest monitoring equipment is in the peak period of rice stem borer in the corresponding generation, and outputs the peak period detection results including the start time of the peak period, the end time of the peak period, the peak date, and the peak intensity. The optimal control window is calculated based on the ecological growth cycle of the stem borer for pest monitoring and early warning. The empirical formula for estimating the optimal prevention window is as follows: Prevention and control window = duration of the previous peak period + ΔT Wherein, ΔT represents the time from the peak period of the previous generation of adult borers to the peak period of the next generation of egg hatching.

[0017] This invention overcomes the limitations of existing methods for detecting peak moth populations in rice stem borers, such as uncorrected outliers, low fitting accuracy, and weak scalability. Starting from the overall variation pattern of insect trapping time-series data, and comprehensively considering the impact of equipment malfunctions and extreme weather conditions on trapping volume, this method effectively suppresses and corrects abnormal interference signals. Furthermore, it utilizes a Gaussian multi-peak model to fit daily time-series data and extract peak period indicators. This method can highlight the dominant trend of insect infestation development against complex noise backgrounds, effectively avoiding interference from local fluctuations in moth peak determination, and improving the objectivity and consistency of moth peak identification results.

[0018] Compared with existing technologies, the beneficial effects of this invention are as follows: By systematically processing anomalies and modeling the overall trend of daily insect attraction data from intelligent monitoring lamps, automatic identification and feature extraction of multiple generations of rice stem borer peaks are achieved without relying on human experience. This method not only adapts to practical application scenarios where the noise levels of insect attraction data vary significantly across different sites and years, but also quantifies and extracts key temporal features of moth peaks, improving the accuracy, stability, and engineering applicability of peak period detection results. It is suitable for regional-scale monitoring and precise control of rice stem borers. Attached Figure Description

[0019] Figure 1 This is a flowchart of a method for detecting peak rice stem borer activity based on data from intelligent monitoring lamps, according to the present invention.

[0020] Figure 2 This is a simplified diagram illustrating the identification and interpolation of outliers in time-series data according to a specific embodiment of the present invention.

[0021] Figure 3 This is a simplified diagram illustrating the smooth fitting of a Gaussian multi-peak curve according to a specific embodiment of the present invention.

[0022] Figure 4 This is a simplified schematic diagram illustrating the comparison of detection results, prevention and control window, and actual information according to a specific embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see Figure 1 A method for detecting peak rice stem borer activity based on data from intelligent monitoring lamps, comprising:

[0025] Step 1, acquisition and preprocessing of pest monitoring time-series data, including: Based on the distribution data of regional pest monitoring stations, raw data on the number of insects attracted by all pest monitoring devices in the target area are collected. Based on device information and date information, time-series data on the number of insects attracted by each device are integrated.

[0026] The raw data on the number of insects attracted was collected on-site by the intelligent monitoring lamp, including the device number, date, and daily number of insects attracted; the pest monitoring equipment is an automated pest trapping device deployed in the area, and each pest monitoring device corresponds to a unique spatial location.

[0027] Step 2, identification and repair of abnormal trapping data based on the operating characteristics of the insect-attracting device, including: An adaptive thresholding method was used to identify abnormally low values ​​in the time-series data of insect attraction. A sliding statistic and a continuous trough exclusion logic were combined to distinguish between abnormal operating conditions of the monitoring lamps and biological fluctuations in the borer population. For all identified abnormally low value intervals, cubic spline interpolation and boundary constraint repair were used to generate an interpolated insect attraction time-series curve. The adaptive thresholding method is as follows:

[0028] The adaptive threshold is dynamically calculated based on the median, 25th percentile, and annual maximum value of growing season data, using the following formula: Adaptive Threshold = max(α·global_max, β·global_median, global_q25, T) min );

[0029] In the formula, `max` represents the maximum value among the four indicators described later; `global_max` is the global maximum value, `global_median` is the median of the growing season, `global_q25` is the 25th percentile of the occurrence season, and `T`... min α is the minimum fixed threshold, β is the global maximum adjustment coefficient, and β is the growing season median adjustment coefficient.

[0030] To preserve the original data trend and avoid introducing spurious peaks during anomaly repair, the 90th quantiles of the data from 30 days before and after the anomaly interval are connected, and then Loess smoothing is used to form the upper envelope. For any time t within the anomaly interval, the interpolated repair value yinterp(t) must satisfy:

[0031] yinterp(t)≤γ·yupper(t)

[0032] Where yinterp(t) is the repair value obtained by cubic spline interpolation at time t; yupper(t) is the value of the upper envelope at time t; γ is the upper envelope relaxation coefficient, and it is recommended that the value range be 1.1≤γ≤1.3.

[0033] Step 3, Construction of the multi-peak Gaussian model, including: SG filtering was used to smooth the repaired insect attraction time-series curve, and potential peak values ​​were initially detected based on the smoothed curve. A multi-peak Gaussian model was constructed, and the model parameters were initialized based on the initial peak values. The model parameters were optimized through curve fitting to generate a fitting curve that highly matches the measured data. The formula for the multi-peak Gaussian model is as follows:

[0034]

[0035] Where t is the time variable, representing the cumulative number of days within the monitoring period; y(t) is the daily insect trapping amount obtained by fitting the Gaussian multi-peak model on day t; n is the number of peaks, p i Let exp represent the intensity of the i-th peak, and let t be the natural exponential function. 0i For peak time, w i The peak width is defined as follows: The multi-peak Gaussian model is composed of multiple Gaussian components superimposed on each other, with each component corresponding to a pest occurrence peak. The model parameters are optimized by fitting using the least squares method to ensure that the mean square error between the fitted curve and the measured data is minimized.

[0036] Step 4, Model Adaptive Optimal Parameter Selection and Peak Detection, includes: For the time series curve of insect attraction, a candidate set of SG filter parameters is set; for each parameter combination, the model building process in step 3 is executed to calculate the determination coefficient R² of Gaussian fitting, the Bayesian information criterion BIC, and the peak position stability index peakvar for each combination; the optimal parameter combination is screened through a weighted scoring mechanism to determine the optimal fitting result, and the peak index of the time series data, i.e. the peak period of adult borer occurrence, is output according to the Gaussian multi-peak model under the optimal parameters.

[0037] The formulas for calculating the coefficient of determination R², the Bayesian information criterion BIC, and the peak position stability index peakvar are as follows:

[0038] Coefficient of determination R2 :

[0039]

[0040]

[0041]

[0042] Where N is the number of samples, y obs For the original observation value, y fit These are Gaussian fitted values. RSS is the arithmetic mean of all original observations; RSS is the residual sum of squares, which is the sum of squares of the differences between the original observations and the Gaussian fitted values; SST is the total sum of squares, which is the sum of squares of the differences between the original observations and the mean of the original observations.

[0043] Bayesian Information Criterion (BIC):

[0044]

[0045] Peak position stability index peakvar:

[0046]

[0047] Where n is the number of peaks; k is the total number of independent parameters of the model, and each Gaussian fitting peak contains 3 Gaussian parameters, k=3n; , These represent the positions of the i-th peak and the (i-1)-th peak after sorting by time, respectively.

[0048] Indicator normalization score: The coefficient of determination R 2 After normalization using the Bayesian Information Criterion (BIC) and the peak position stability index (peakvar), a final score is synthesized according to weights, and the window with the highest score is selected as the optimal parameter.

[0049]

[0050] Among them, w1, w2, and w3 are the weight coefficients of the Bayesian Information Criterion (BIC), the Coefficient of Determination (R²), and the Peak Position Stability Index (peakvar), respectively.

[0051] Step 5, peak merging and ecological generation classification, includes: Based on the fitting results corresponding to the optimal parameters, each local peak is extracted, and the intensity, peak time, peak width, and corresponding 95% time confidence interval of each peak are calculated. By performing overlap checks on the time confidence intervals of different peaks, redundant peaks caused by noise or overfitting are filtered out, and statistically significant effective peaks are retained. For the selected effective peaks, generation matching is performed in conjunction with the occurrence pattern of rice stem borer.

[0052] The peak selection logic is based on a 95% confidence interval overlap check. If the time confidence intervals of two peaks overlap, the peak with higher intensity is retained. The generation label assignment adds artificial ecological constraints, and the final label is determined by the matching degree between the peak time and the generation range division.

[0053] Step 6, Peak Period Information Extraction and Prevention Window Output, including: When at least one valid peak is detected within the monitoring period, the pest monitoring equipment is determined to be in the peak period of the corresponding generation of rice stem borer. The system outputs peak period detection results including the start time, end time, peak date, and peak intensity. Based on the stem borer's ecological growth cycle, the optimal control window is calculated for pest monitoring and early warning. Prevention and control window = duration of the previous peak period + ΔT Wherein, ΔT represents the time from the peak period of the previous generation of adult borers to the peak period of the next generation of egg hatching.

[0054] Example: This example uses Yuhang District, Hangzhou City, Zhejiang Province as a typical area and the rice stem borer (Chilo suppressalis) as a typical example. This area has a typical rice-growing ecological environment. The monitoring data comes from the insect-attracting monitoring data of intelligent monitoring lamps in four towns (streets) within Yuhang District, including 13 monitoring points in Yuhang Street, Cangqian Street, Jingshan Town, and Pingyao Town. Data from monitoring lamp No. 447015 in Lianjutang Village, Cangqian Street, is used as a typical sample to demonstrate the specific implementation process of this invention.

[0055] The four town (street) monitoring points cover the typical rice cultivation distribution and geographical environment differences in Yuhang District, and can represent the insect pest change characteristics of the main rice field ecological zones in the district.

[0056] The implementation process of this invention includes the following steps:

[0057] Step 1, Acquisition and Preprocessing of Pest Monitoring Time Series Data: Based on the distribution information of regional pest monitoring stations, the original time series of rice stem borer attractant quantities from 13 monitoring points in four towns (streets) of Yuhang District from 2022 to 2024 was obtained. The original data includes fields such as equipment number, date, and attractant quantity, and was collected daily by an automated pest monitoring device.

[0058] Data from each device were aggregated in chronological order, standardized in format, and missing values ​​were handled to construct a time-series database of insect attraction in the region, providing raw input for subsequent analysis.

[0059] Step 2, Identification and Repair of Abnormal Trapping Data Based on the Operating Characteristics of the Insect-Trapping Device: For abnormally low values ​​caused by abnormal operating conditions of the monitoring lamps or biological processes of adult borers, this embodiment uses an adaptive threshold method for abnormal low value detection. The dynamic threshold calculation formula is shown below:

[0060] Adaptive threshold = max(α·global_max, β·global_median, global_q25, T) min )

[0061] Where global_max is the maximum number of insects attracted throughout the year, global_median is the median during the growing season (March–October), global_q25 is the 25th percentile, and T min The minimum fixed threshold is set here as α=0.05, β=0.3, and T. min =3.

[0062] The detected abnormal intervals were repaired using cubic spline interpolation, and upper envelope and peak width constraints were set to avoid the introduction of spurious peaks. The data curve after repair for monitoring light number 447015 in 2023 is shown below. Figure 2 As shown, it can clearly interpolate abnormally low value ranges caused by equipment failure or abnormal weather, and the true peak shape is effectively restored.

[0063] Step 3, Multi-peak Gaussian Model Construction: The repaired insect-attracting time-series data were preprocessed using Savitzky-Golay (SG) filtering to smooth and preliminarily identify potential peaks. The fitting results for monitoring lamp No. 447015 in 2023 are as follows... Figure 3 As shown, a multi-peak Gaussian model is constructed using the identified local peaks as the initial points, and the formula is as follows:

[0064]

[0065] Step 4, Model Adaptive Optimal Parameter Selection and Peak Detection: For different measurement point data characteristics and noise levels, this invention sets a candidate set of SG filter parameters (window length, order combination). Gaussian fitting is performed on each set of parameters, and the index is calculated. The index calculation formula is shown below:

[0066]

[0067]

[0068]

[0069] in, , y obs For the original observation value, y fit represents the Gaussian fitted value, and N is the number of samples; These represent the positions of the i-th peak and the (i-1)-th peak after sorting by time, respectively.

[0070] After normalizing each indicator, the coefficient of determination R... 2 After normalization using the Bayesian Information Criterion (BIC) and the peak position stability index (peakvar), a final score is synthesized according to weights, and the window with the highest score is selected as the optimal parameter.

[0071]

[0072] Where w1, w2, and w3 are the weight coefficients of the Bayesian Information Criterion (BIC), the Coefficient of Determination (R²), and the Peak Position Stability Index (peakvar), respectively, with default weights of w1=0.4, w2=0.4, and w3=0.2. The parameter group with the highest overall score is selected as the optimal parameter combination, and the corresponding fitted curve and peak period identification result are used as the final model output.

[0073] Step 5, Peak Merging and Ecological Generation Classification: Under optimal parameters, extract the intensity, time, width, and 95% confidence interval of each peak. For peaks with overlapping time confidence intervals, retain those with higher intensity to eliminate fitting redundancy. Combining the biological occurrence patterns of the rice stem borer and regional meteorological characteristics, classify the peaks according to their time range, such as the first generation, second generation, third generation, and overwintering generation.

[0074] Step 6, Peak Period Information Extraction and Prevention Window Output: Summarize the peak period indicators and generations from the above steps, and output the final peak information obtained from the monitoring light at No. 447015, Lianjutang Village, Cangqian Street, as shown in Table 1.

[0075] Table 1 Peak Value Information

[0076] The peak value distribution of 13 monitoring lights in four towns (subdistricts) of Yuhang District is as follows: Figure 4 As shown in the figure, the green area represents the main peak occurrence periods of adult rice stem borers over the years, as published by the Yuhang District Plant Protection Station based on manual field surveys, while the gray area represents the estimated control window. The figure demonstrates that the peak period results automatically extracted by this method are highly consistent with the manual monitoring results, accurately identifying the occurrence time of each generation's peak.

[0077] Because this invention is based on daily continuous monitoring data from intelligent monitoring lamps, it can achieve uninterrupted automatic detection and calculation throughout the year. Therefore, while ensuring accuracy, it significantly improves the timeliness and spatial coverage of the data. Furthermore, the automated identification results can not only reproduce the peak period determined by manual surveys, but also further reveal subtle differences between different monitoring points within the same area, providing more valuable quantitative evidence for refined forecasting and regionalized control. Compared to traditional methods relying on manual experience and phased observations, this invention enables data-driven objective identification, automated processing, and cross-year comparative analysis. In practical applications, the peak period detection results of this invention can be combined with the stem borer growth and development cycle to estimate different control windows, providing a basis for scientific control.

[0078] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for detecting the peak period of rice stem borer based on data from intelligent monitoring lamps, characterized in that, include: Step 1: Acquire and preprocess the time-series data of insect attraction quantity; Step 2: Identify and repair abnormal trapping data based on the operating characteristics of the insect-attracting device; Step 3: Construct a multi-peak Gaussian model; Step 4: Perform adaptive optimal parameter selection and peak detection for the multi-peak Gaussian model; Step 5: Perform peak merging and ecological generation classification; Step 6: Extract peak period information and output prevention and control information.

2. The method for detecting peak rice stem borer activity based on intelligent monitoring lamp data according to claim 1, characterized in that, Step 1 includes: Based on the distribution data of regional pest monitoring stations, raw data on the number of insects attracted by all pest monitoring devices in the target area were collected and integrated to obtain time-series data on the number of insects attracted by each device.

3. The method for detecting peak rice stem borer activity based on intelligent monitoring lamp data according to claim 1, characterized in that, Step 2 includes: An adaptive threshold method was used to identify abnormally low values ​​in the time series data of insect attraction, and to distinguish between abnormal operating conditions of the monitoring lamps and biological fluctuations of the borers. For all identified abnormally low value intervals, cubic spline interpolation and boundary constraint repair were used to generate the time series curve of insect attraction after abnormally low value interpolation.

4. The method for detecting peak rice stem borer activity based on intelligent monitoring lamp data according to claim 3, characterized in that, The adaptive threshold method in step 2 is as follows: The adaptive threshold is dynamically calculated based on the median, 25th percentile, and annual maximum value of growing season data, using the following formula: Adaptive Threshold = max(α·global_max, β·global_median, global_q25, T) min ); In the formula, `max` represents the maximum value among the four indicators described later; `global_max` is the global maximum value, `global_median` is the median of the growing season, `global_q25` is the 25th percentile of the occurrence season, and `T`... min The minimum fixed threshold is α, the global maximum adjustment coefficient is β, and the median adjustment coefficient is β. To preserve the original data trend and avoid introducing spurious peaks during anomaly repair, the 90th quantiles of the data from 30 days before and after the anomaly interval are connected, and then Loess smoothing is used to form the upper envelope. For any time t within the anomaly interval, the interpolated repair value yinterp(t) must satisfy: yinterp(t)≤ γ·yupper(t) Where yinterp(t) is the repair value obtained by cubic spline interpolation at time t; yupper(t) is the value of the upper envelope at time t; and γ is the upper envelope relaxation coefficient.

5. The method for detecting peak rice stem borer activity based on intelligent monitoring lamp data according to claim 3, characterized in that, Step 3 includes: The time-series curve of insect attraction after repair was smoothed by SG filtering, and potential peak values ​​were initially detected based on the smoothed curve. A multi-peak Gaussian model was constructed, and the model parameters were initialized based on the initial peak values. The model parameters were optimized by curve fitting to generate a fitting curve that closely matches the measured data. The formula for the Gaussian multi-peak model is as follows: ， Where t is the time variable, representing the cumulative number of days within the monitoring period; y(t) is the daily insect trapping amount obtained by fitting the Gaussian multi-peak model on day t; n is the number of peaks, p i Let exp represent the intensity of the i-th peak, and let t be the natural exponential function. 0i For peak time, w i The peak width is defined as follows: The multi-peak Gaussian model is composed of multiple Gaussian components superimposed on each other, with each component corresponding to a pest occurrence peak. The model parameters are optimized by fitting using the least squares method to ensure that the mean square error between the fitted curve and the measured data is minimized.

6. The method for detecting peak rice stem borer activity based on intelligent monitoring lamp data according to claim 5, characterized in that, Step 4 includes: For the time series curve of insect attraction, a candidate set of SG filter parameters is set; for each parameter combination, the model building process in step 3 is executed to calculate the determination coefficient of Gaussian fitting, Bayesian information criterion and peak position stability index for each combination, and the parameter combination with the best overall performance is selected.

7. The method for detecting peak rice stem borer activity based on intelligent monitoring lamp data according to claim 6, characterized in that, In step 4, the coefficient of determination R 2 The Bayesian Information Criterion (BIC) and the peak position stability index (peakvar) are calculated as follows: Coefficient of determination R 2 : ， ， ， Where N is the number of samples, y obs For the original observation value, y fit These are Gaussian fitted values. RSS is the arithmetic mean of all original observations; RSS is the residual sum of squares, which is the sum of squares of the differences between the original observations and the Gaussian fitted values; SST is the total sum of squares, which is the sum of squares of the differences between the original observations and the mean of the original observations. Bayesian Information Criterion (BIC): ， Peak position stability index peakvar: ， Where n is the number of peaks; k is the total number of independent parameters of the model, and each Gaussian fitting peak contains 3 Gaussian parameters, k=3n; , These represent the positions of the i-th peak and the (i-1)-th peak after sorting by time, respectively. Indicator normalization score: The coefficient of determination R 2 After normalization using the Bayesian Information Criterion (BIC) and the peak position stability index (peakvar), a final score is synthesized according to weights, and the window with the highest score is selected as the optimal parameter. ， Among them, w1, w2, and w3 are the weight coefficients of the Bayesian Information Criterion (BIC), the Coefficient of Determination (R²), and the Peak Position Stability Index (peakvar), respectively.

8. The method for detecting peak rice stem borer activity based on intelligent monitoring lamp data according to claim 6, characterized in that, Step 5 includes: Based on the fitting results corresponding to the optimal parameters, each local peak is extracted, and the intensity, peak time, peak width, and corresponding 95% time confidence interval of each peak are calculated. By performing overlap checks on the time confidence intervals of different peaks, redundant peaks caused by noise or overfitting are filtered out, and statistically significant effective peaks are retained. For the selected effective peaks, generation matching is performed in conjunction with the occurrence pattern of rice stem borer.

9. A method for detecting peak rice stem borer activity based on data from intelligent monitoring lamps according to claim 8, characterized in that, The peak selection logic in step 5 is based on the 95% confidence interval overlap check. If the time confidence intervals of two peaks overlap, the peak with higher intensity is retained. The generation label assignment adds artificial ecological constraints, and the final label is determined by the matching degree between the peak time and the generation range division.

10. The method for detecting peak rice stem borer activity based on intelligent monitoring lamp data according to claim 8, characterized in that, Step 6 includes: When at least one valid peak is detected within the monitoring period, it is determined that the pest monitoring equipment is in the peak period of rice stem borer in the corresponding generation, and outputs the peak period detection results including the start time of the peak period, the end time of the peak period, the peak date, and the peak intensity. The optimal control window is calculated based on the ecological growth cycle of the stem borer for pest monitoring and early warning. The empirical formula for estimating the optimal prevention window is as follows: Prevention and control window = duration of the previous peak period + ΔT Wherein, ΔT represents the time from the peak period of the previous generation of adult borers to the peak period of the next generation of egg hatching.

Images