Farmland insect pest real-time monitoring and early warning method and device based on acoustic detection

Abstract

The invention provides a farmland insect pest real-time monitoring and early warning method and device based on acoustic detection, and the method comprises the steps: collecting a sound signal in a farmland through a microphone array, carrying out the denoising of the sound signal through an audio preprocessing unit, extracting a feature vector of the sound signal, and transmitting the feature vector to a local data processing unit; the local data processing unit compares the feature vectors of the sound signals with a standard data set by adopting a classification algorithm to obtain insect type and frog type information corresponding to the sound signals, carries out sound signal sampling according to the insect type and frog type information, determines the number of insects and frogs, and sends the number of insects and frogs to the central data processing unit; the central data processing unit sets an early warning threshold value according to the number of frogs and insects, and reminds a user of deinsectization when the early warning threshold value is exceeded. According to the invention, sound source detection and big data analysis are combined, accurate monitoring and early warning of farmland insect pests are realized, the scheme cost is low, pesticide use can be effectively reduced, and the ecological environment is protected while the product safety is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of pest control, and in particular to a method and device for real-time monitoring and early warning of farmland pests based on acoustic detection. Background Art

[0002] Pests and diseases are a global agricultural problem that has long caused huge losses to agriculture and is a challenge in agricultural production. Pests and diseases not only directly affect the yield and quality of crops, but can also cause price fluctuations of agricultural products and threaten food security.

[0003] Most of the existing pest control methods are biological control, physical control and chemical control. Biological control reduces the number of pests by introducing natural enemies corresponding to pests, but the cost is high and small-scale farmers cannot afford it, and the insecticidal effect is slow and not suitable for rapid control. Physical control uses physical or mechanical methods to control pests, such as using light to drive away, hot water soaking, high temperature treatment and setting nets. Although physical control does not involve chemicals and is therefore environmentally friendly, its large-scale pest control efficiency is very low and is only suitable as a temporary auxiliary control measure. Chemical control kills pests in farmland by spraying pesticides, which is efficient and easy to implement. It is the current mainstream pest control method. At present, the application methods of chemical control are divided into general spraying, precision spraying, spot spraying, strip spraying, etc. General spraying causes serious pollution to the environment, and will lead to the rapid development of pest resistance, and will also cause waste of resources. Compared with other application methods, precision spraying saves the most pesticide use and has the least pollution to the environment. In order to achieve precise spraying, the current mainstream recognition method is to combine computer vision, artificial intelligence, machine learning algorithms and other technologies. However, due to the high-priced hardware involved, the cost of this technology is high, especially for small and medium-sized farmers. Even if the subsequent returns are greater than the investment, there is not enough money to deploy this technology in the early stage. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a method and device for real-time monitoring and early warning of farmland insect pests based on acoustic detection with low cost and high recognition rate.

[0005] Technical solution: A real-time monitoring and early warning method for farmland pests based on acoustic detection, comprising the following steps:

[0006] (1) Collecting sound signals in the target area;

[0007] (2) De-noising the sound signal and extracting the feature vector of the sound signal;

[0008] (3) using a classification algorithm to compare the feature vector of the sound signal with a standard data set to obtain the insect species information and frog species information corresponding to the sound signal; sampling the sound signal according to the insect species information and frog species information to determine the insect quantity information and frog quantity information;

[0009] (4) According to the frog population information and insect population information, set the warning threshold and issue an alarm when the warning threshold is exceeded.

[0010] Specifically, step (2) includes:

[0011] (21) Discretize the sound signal A(t) to obtain A(n), and calculate the signal frequency based on A(n):

[0012]

[0013] Where: A(t) represents the sound pressure level at each time point, n in A(n) is the index of the sampling point, the total number of sampling points is N, and f k is the signal frequency, f s is the sampling frequency, which is set when collecting sound signals, and k is the index of the sampling point in the frequency domain;

[0014] (22) The sound signal is denoised using a filter, and the frequency response formula of the filter is:

[0015]

[0016] Where: H(f k ) is the frequency response of the filter, f k is the signal frequency, f c is the cut-off frequency, n is the order of the filter;

[0017] (23) After being processed by the filter, the denoised sound signal A′(t) is obtained. The denoised sound signal A′(t) is subjected to a fast Fourier transform to convert the time domain signal into a frequency domain signal:

[0018]

[0019] Where: B(f) is the frequency domain signal, A′(t) is the denoised time domain signal, f is the frequency, t is the time, and j is the imaginary unit;

[0020] (24) The main frequency components and amplitude information are extracted from the frequency domain signal B(f) to obtain the amplitude spectrum M(f) and form the feature vector C, which is as follows:

[0021] M(f)=|B(f)|

[0022] C=[M(f 1 ),M(f 2),……,M(f n )]

[0023] Where: M(f) is the amplitude spectrum, f 1 ,f 2 ,…,f n is the main frequency component.

[0024] Preferably, the above filter is a fourth-order filter.

[0025] Specifically, the classification algorithms include support vector machine algorithm, random forest algorithm, decision tree algorithm, K nearest neighbor algorithm, artificial neural network algorithm, and naive Bayes algorithm.

[0026] Preferably, in step (3), when the classification algorithm is a K nearest neighbor algorithm, using the classification algorithm to compare the feature vector of the sound signal with the standard data set to obtain the insect species information and frog species information corresponding to the sound signal includes the following steps:

[0027] (311) Each eigenvalue of the eigenvector C of the sound signal is normalized:

[0028]

[0029] Where: is the jth eigenvalue of the eigenvector C after normalization, x j represents the jth eigenvalue of eigenvector C;

[0030] (312) Extract the feature vectors of various insect or frog samples in the standard data set, perform the normalization process in step (311), and calculate the Euclidean distance between the feature vector C of the sound signal after the normalization process and the feature vector of various insect or frog samples in the standard data set:

[0031]

[0032] Where: d i is the Euclidean distance between C and the i-th eigenvector in the standard dataset, C j is the jth eigenvalue of the eigenvector in the standard data set, is the jth eigenvalue of the normalized eigenvector C, and D is the dimension of the eigenvector;

[0033] (313) Arrange the Euclidean distances between all feature vectors in the standard data set and feature vector C in order;

[0034] (314) Given a k value, select the k eigenvectors with the smallest Euclidean distance to the eigenvector C and put the corresponding k sample indexes into the set I neighbours ={i|b i}, bi is one of the k samples with the smallest Euclidean distance, and then the insect species or frog species corresponding to the k samples are put into the set E = {e i |i∈I neighbours}, e i is the insect species or frog species corresponding to the i-th sample;

[0035] (315) According to the insect species or frog species in the set E, the insect species or frog species corresponding to the feature vector C is determined by majority voting to obtain the insect species information or frog species information:

[0036]

[0037] Where: is the indicator function, when c i When =c, the result is 1, otherwise it is 0.

[0038] Specifically, in step (3), sampling the sound signal according to the insect species information and the frog species information, and determining the insect quantity information and the frog quantity information includes the following steps:

[0039] (321) According to the information of insect species or frog species, the average time T for the insect or frog to call once is taken, and the sound signal A(t) is collected at intervals T, and a total of M times is collected to obtain the discrete amplitude s i :

[0040] s i =A(t i ), i∈[1,M]

[0041] Where: A(t i ) is the sound signal collected for the i-th time;

[0042] (322) The sound signal s i Converted into digital signal s′ i :

[0043] s′ i =ADC(s i )

[0044] (323) According to the digital signal s′ i Calculate the energy E k :

[0045]

[0046] (324) Set the energy threshold. If the energy E k If it is greater than the energy threshold, it is determined as an event, and the total number of events in M ​​time periods T is calculated to obtain the corresponding insect quantity information or frog quantity information.

[0047] Specifically, step (4) includes:

[0048] When the number of frogs in the area is less than the set value, the warning threshold is set to 1 / 4 of the insect population carrying capacity K. When the warning threshold is exceeded, a warning message is sent to the user;

[0049] When the number of frogs in the area is greater than or equal to the set value, an early warning threshold is set according to the number of frogs, and the early warning threshold is greater than 1 / 4K and less than 1 / 2K.

[0050] Preferably, step (4) further includes: calculating the insect density of the target area in different time periods according to the insect quantity information, obtaining the time period with the highest insect density by statistics, and sending warning information to the user.

[0051] The present invention also provides a real-time monitoring and early warning device for farmland insect pests based on acoustic detection, which is characterized in that it includes a plurality of monitoring units and a central data processing unit, the monitoring unit includes a microphone array, an audio preprocessing unit, and a local data processing unit, the microphone array includes a plurality of symmetrically distributed microphones, the microphone array, the audio preprocessing unit, and the local data processing unit are connected in sequence, and the local data processing unit is connected to the central data processing unit;

[0052] Microphone array: used to collect sound signals in the target area and transmit them to the audio preprocessing unit;

[0053] Audio preprocessing unit: used to perform denoising on the sound signal, extract the feature vector of the sound signal, and transmit it to the local data processing unit;

[0054] Local data processing unit: used to compare the characteristic vector of the sound signal with the standard data set by using a classification algorithm to obtain the insect species information and frog species information corresponding to the sound signal; sample the sound signal according to the insect species information and frog species information, determine the insect quantity information and frog quantity information, and transmit them to the central data processing unit;

[0055] Central data processing unit: used to set warning thresholds based on frog population information and insect population information, and to issue an alarm when the warning thresholds are exceeded.

[0056] Specifically, the arrangement of the microphone array includes: uniform grid distribution: each microphone is equidistantly arranged in a rectangular grid; boundary reinforcement distribution: the number of microphones is increased at the boundary of the target area and the number of microphones is reduced at the center; annular distribution: the microphones are arranged in a circle or multiple concentric circles around the center of the target area.

[0057] Beneficial effects: Compared with the prior art, the present invention has the following significant effects:

[0058] (1) By combining acoustic detection with big data analysis technology, the present invention realizes the real-time collection and analysis of information on the types and numbers of insects and frogs in farmland, achieving automatic and efficient information collection.

[0059] (2) By using the information on the types and numbers of insects and frogs collected, it is possible to determine in real time whether the number of pests has reached the warning threshold, thereby sending pest control reminders to farmers in a timely and effective manner. At the same time, accurate judgments can be made in terms of time and region, thereby improving the accuracy and efficiency of pest control, reducing the excessive use of pesticides, contributing to the protection of the ecological environment, and reducing the residual pesticide components in crops.

[0060] (3) The equipment used is relatively low-priced and has cost advantages, making it suitable for use by small and medium-sized farmers. At the same time, the equipment settings are simplified, and a single monitoring unit can be used independently off the grid, making it suitable for a variety of scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flow chart of the method of embodiment 1 of the present invention.

[0062] Figure 2 It is a structural diagram of the device of embodiment 2 of the present invention. DETAILED DESCRIPTION

[0063] The present invention is further explained below in conjunction with the accompanying drawings and specific implementation methods.

[0064] Example 1

[0065] In this embodiment, the standard data sets used are three public data sets: Macaulay Library, GBIF, and AmphibiaWeb, which contain a large amount of sound information of insects and frogs.

[0066] See also Figure 1 As shown, this embodiment provides a method for real-time monitoring and early warning of farmland pests based on acoustic detection, comprising the following steps:

[0067] (1) Collect the sound signal A(t) in the target area, where A(t) represents the sound pressure level at each time point.

[0068] (2) Perform denoising on the sound signal A(t) to remove background noise such as wind and rain, and extract the feature vector of the sound signal, including information such as frequency, loudness, and duration.

[0069] Discretize the sound signal A(t) to get A(n), and calculate the signal frequency based on A(n):

[0070]

[0071] Where: A(t) represents the sound pressure level at each time point, n in A(n) is the index of the sampling point, the total number of sampling points is N, and f k is the signal frequency, f s is the sampling frequency, which is the value set when collecting the signal, and k is the index of the sampling point in the frequency domain.

[0072] Using analog signal processing technology, amplifiers and filters are used to optimize sound signals. Taking insect sound signals as an example, by analyzing the insect sound data in the Macaulay Library, it is found that effective insect sound signals are concentrated in the frequency band range of 2kHz to 8kHz. The filter is used to filter the waves outside the frequency band in the sound signal. In this embodiment, considering efficiency, performance, cost and implementation complexity, a fourth-order filter is selected, with a low cutoff frequency set to 2kHz and a high cutoff frequency set to 8kHz to retain the sound characteristics of insects and filter out low-frequency and high-frequency environmental noise. The frequency response formula of the filter is:

[0073]

[0074] Where: H(f k ) is the frequency response of the filter, f k is the signal frequency, f c is the cut-off frequency, n is the order of the filter;

[0075] After being processed by the filter, the denoised sound signal A′(t) is obtained. The denoised sound signal A′(t) is subjected to fast Fourier transform (FFT) to convert the time domain signal into a frequency domain signal:

[0076]

[0077] Where: B(f) is the frequency domain signal, A′(t) is the denoised time domain signal, f is the frequency, t is the time, and j is the imaginary unit;

[0078] Extract the main frequency components and amplitude information from the frequency domain signal B(f), take the amplitude modulus to obtain the amplitude spectrum M(f), and form the feature vector C. The formula is as follows:

[0079] M(f)=|B(f)|

[0080] C=[M(f 1 ),M(f 2 ),……,M(f n )]

[0081] Where: M(f) is the amplitude spectrum, f 1 ,f 2 ,…,f n is the main frequency component.

[0082] (3) Using a classification algorithm, the characteristic vector of the sound signal is compared with a standard data set to obtain the insect species information and frog species information corresponding to the sound signal; based on the insect species information and frog species information, the sound signal is sampled to determine the insect quantity information and frog quantity information.

[0083] In the task of processing sound signal data classification, common classification algorithms include support vector machine algorithm (SVM), random forest algorithm, decision tree algorithm, K nearest neighbor algorithm (KNN), artificial neural network algorithm, naive Bayes algorithm, etc. In this embodiment, combined with the data set and test sample situation used, the K nearest neighbor algorithm is simple and easy to implement, has a parameter-free model, is suitable for small data sets, can be dynamically updated, and is consistent with the use scenario of this method. The following describes the practical application of the K nearest neighbor algorithm in this method.

[0084] Normalize each eigenvalue of the eigenvector C of the sound signal:

[0085]

[0086] Where: is the jth eigenvalue of the eigenvector C after normalization, x j Represents the j-th eigenvalue of eigenvector C.

[0087] In order to distinguish the types of insects and frogs, the feature vectors of various insect or frog samples in the standard data set are extracted and normalized. The Euclidean distance between the feature vector C of the normalized sound signal and the feature vector of various insect or frog samples in the standard data set is calculated:

[0088]

[0089] Where: d i is the Euclidean distance between C and the i-th eigenvector in the standard dataset, C j is the jth eigenvalue of the eigenvector in the standard data set, is the j-th eigenvalue of the normalized eigenvector C, and D is the dimension of the eigenvector.

[0090] Arrange the Euclidean distances between all feature vectors in the standard data set and feature vector C in order;

[0091] Given a k value, which is generally determined by the size of the target area and the environment, select the k feature vectors with the smallest Euclidean distance to the feature vector C and put the corresponding k sample indexes into the set I neighbours ={i|b i}, and then put the insect species or frog species corresponding to the k samples into the set E = {e i |i∈Ineighbours} in; b i is one of the k data samples with the smallest Euclidean distance, e i is the insect species or frog species corresponding to the i-th data sample.

[0092] In this embodiment, the k value in the K nearest neighbor algorithm is determined as follows: the samples in the standard data set are divided into ten groups, one of which is taken for testing each time, and the remaining nine groups are used as samples. The steps are the same as described above, and the accuracy obtained by the voting method is compared. Finally, the average value of the ten groups is taken, and the k value with the highest average accuracy is used as the k value for actual application.

[0093] According to the insect species or frog species in set E, the majority voting method is used to obtain the insect species or frog species with the most votes as the output result, thereby determining the insect species or frog species corresponding to the feature vector C and obtaining the insect species information or frog species information:

[0094]

[0095] Where: is the indicator function, when c i When =c, the result is 1, otherwise it is 0.

[0096] According to the information of insect species or frog species, the average time T for the insect or frog to call once is taken, and the sound signal A(t) is collected at intervals of T, and a total of M times is collected to obtain the discrete amplitude s i :

[0097] s i =A(t i ), i∈[1,M]

[0098] Where: A(t i ) is the sound signal collected for the i-th time.

[0099] The sound signal s i Converted into digital signal s′ i :

[0100] s′ i =ADC(s i )

[0101] According to the digital signal s′ i Calculate the energy E k :

[0102]

[0103] Set the energy threshold E thresh , if the energy E k Greater than the energy threshold E thresh, it is determined as an event, and the total number of events in M ​​time periods T is calculated. It can be assumed that the total number of events represents the number of insects or frogs in the target area. Repeat the above steps to obtain all the insect population information or frog population information in the target area.

[0104] (4) According to the frog population information and insect population information, combined with the insect population carrying capacity, an early warning threshold is set. When the early warning threshold is exceeded, an alarm is issued to remind the user to perform pest control operations.

[0105] The number of frogs is an important reference factor for the extent of insect infestation. Since the predatory behavior of frogs has a natural regulatory effect on the insect population, the impact of frogs should be taken into account when judging the extent of insect infestation.

[0106] When the number of frogs in the area is less than the set value, such as when the number of frogs is 0, its impact is not considered. At this time, the insect population carrying capacity (K value) is considered, that is, the maximum number of insects that can be maintained under the support of existing resources and environmental conditions in a specific environment. When the population reaches K / 2, the insect growth rate is the fastest and difficult to control, so it is necessary to control before the population reaches K / 2. In this embodiment, the warning threshold is set to K / 4. When the warning threshold is exceeded, a warning message is sent to the user to remind the user to take pest control measures in time;

[0107] When the number of frogs in the area is greater than or equal to the set value, such as exceeding 10% of the number of insects, an early warning threshold is set according to the number of frogs. The early warning threshold is greater than 1 / 4K and less than 1 / 2K. If it reaches 10% of the number of insects, the early warning threshold is adjusted to K / 3.

[0108] This embodiment also provides an additional early warning method, which calculates the insect density in the target area in different time periods based on the insect quantity information, draws a curve of the insect density changing over time, statistically obtains the time period with the highest insect density, sends early warning information to the user, and recommends that the user take pest control measures in the corresponding time period.

[0109] Example 2

[0110] Please refer to Figure 2 As shown, this embodiment provides a real-time monitoring and early warning device for farmland insect pests based on acoustic detection, which includes a plurality of monitoring units and a central data processing unit. The monitoring unit includes a microphone array, an audio preprocessing unit, and a local data processing unit. The monitoring unit is arranged in a target area, i.e., a farmland. The microphone array, the audio preprocessing unit, and the local data processing unit are connected in sequence via wired / wireless means, and all local data processing units are connected to the central data processing unit via wired / wireless means.

[0111] Microphone array: used to collect sound signals in the target area and transmit them to the audio preprocessing unit.

[0112] In this embodiment, a planar array structure is used to design the microphone array. A microphone array includes multiple microphones, which are symmetrically distributed on the array plane to ensure effective processing of sound sources in all directions. The structure of the planar array is much simpler than that of the spherical array and the cost is lower, but the performance is sufficient to meet the needs of detecting insect and frog sounds in farmland environments.

[0113] In this embodiment, the arrangement of the microphone array includes: uniform grid distribution: each microphone is equidistantly arranged in a rectangular grid. This distribution helps the system to evenly cover the entire farmland and simplifies the complexity of sound source positioning; boundary reinforcement distribution: the number of microphones is increased at the boundary of the target area and the number of microphones is reduced at the center. Since the boundary area is the primary area for insects to invade farmland, this distribution can better capture the sounds of insects entering the farmland from the outside; ring distribution: the microphones are arranged in a circle or multiple concentric circles around the center of the target area. This distribution is suitable for farmland areas that are close to a circle.

[0114] Audio preprocessing unit: used to denoise the sound signal, extract the feature vector of the sound signal, and transmit it to the local data processing unit.

[0115] Local data processing unit: used to compare the characteristic vector of the sound signal with the standard data set using a classification algorithm to obtain the insect species information and frog species information corresponding to the sound signal; sample the sound signal according to the insect species information and frog species information, determine the insect quantity information and frog quantity information, and transmit them to the central data processing unit.

[0116] The local data processing unit only needs to process the data from the microphone array where it is located. The computing requirements are low and can be met by running offline using low-cost hardware. This design reduces dependence on real-time network connections and ensures stable operation in remote areas with poor network coverage.

[0117] Central data processing unit: used to set warning thresholds based on frog population information and insect population information, combined with the insect population carrying capacity. When the warning threshold is exceeded, a message can be sent to the user in real time to remind the user to perform pest control operations.

[0118] The central data processing unit collects and integrates the processing results of each local data processing unit, and performs global analysis and decision support. The algorithm flow designed in the present invention requires fewer computing resources, has a low workload, and has low hardware requirements, which can effectively save costs. At the same time, the central data processing unit can also regularly upload data to the cloud for backup and further analysis.

Claims

1. A method for real-time monitoring and early warning of farmland insect pests based on acoustic detection, characterized in that: The following steps are involved: (1) Collecting sound signals in the target area; (2) De-noising the sound signal and extracting the feature vector of the sound signal; (3) using a classification algorithm to compare the feature vector of the sound signal with a standard data set to obtain the insect species information and frog species information corresponding to the sound signal; sampling the sound signal according to the insect species information and frog species information to determine the insect quantity information and frog quantity information; (4) According to the frog population information and insect population information, set the warning threshold and issue an alarm when the warning threshold is exceeded.

2. The method for real-time monitoring and early warning of farmland insect pests according to claim 1, characterized in that: The step (2) comprises: (21) Discretize the sound signal A(t) to obtain A(n), and calculate the signal frequency based on A(n): Where: A(t) represents the sound pressure level at each time point, n in A(n) is the index of the sampling point, the total number of sampling points is N, and f k is the signal frequency, f s is the sampling frequency, which is set when collecting sound signals, and k is the index of the sampling point in the frequency domain; (22) The sound signal is denoised using a filter, and the frequency response formula of the filter is: Where: H(f k ) is the frequency response of the filter, f k is the signal frequency, f c is the cut-off frequency, n is the order of the filter; (23) After being processed by the filter, the denoised sound signal A′(t) is obtained. The denoised sound signal A′(t) is subjected to a fast Fourier transform to convert the time domain signal into a frequency domain signal: Where: B(f) is the frequency domain signal, A′(t) is the denoised time domain signal, f is the frequency, t is the time, and j is the imaginary unit; (24) The main frequency components and amplitude information are extracted from the frequency domain signal B(f) to obtain the amplitude spectrum M(f) and form the feature vector C, which is as follows: M(f)=|B(f)| C=[M(f1),M(f2),……,M(f n )] Where: M(f) is the amplitude spectrum, f1,f2,…,f n is the main frequency component.

3. The method for real-time monitoring and early warning of farmland insect pests according to claim 2, characterized in that: The filter is a fourth-order filter.

4. The method for real-time monitoring and early warning of farmland insect pests according to claim 1, characterized in that: The classification algorithms include support vector machine algorithm, random forest algorithm, decision tree algorithm, K nearest neighbor algorithm, artificial neural network algorithm, and naive Bayes algorithm.

5. The method for real-time monitoring and early warning of farmland insect pests according to claim 1, characterized in that: In the step (3), when the classification algorithm is the K nearest neighbor algorithm, using the classification algorithm to compare the feature vector of the sound signal with the standard data set to obtain the insect species information and frog species information corresponding to the sound signal includes the following steps: (311) Each eigenvalue of the eigenvector C of the sound signal is normalized: Where: is the jth eigenvalue of the eigenvector C after normalization, x j represents the jth eigenvalue of eigenvector C; (312) Extract the feature vectors of various insect or frog samples in the standard data set, perform the normalization process in step (311), and calculate the Euclidean distance between the feature vector C of the sound signal after the normalization process and the feature vector of various insect or frog samples in the standard data set: Where: d i is the Euclidean distance between C and the i-th eigenvector in the standard dataset, C j is the jth eigenvalue of the eigenvector in the standard data set, is the jth eigenvalue of the normalized eigenvector C, and D is the dimension of the eigenvector; (313) Arrange the Euclidean distances between all feature vectors in the standard data set and feature vector C in order; (314) Given a k value, select the k eigenvectors with the smallest Euclidean distance to the eigenvector C and put the corresponding k sample indexes into the set I neighbours ={i|b i }, b i is one of the k samples with the smallest Euclidean distance, and then the insect species or frog species corresponding to the k samples are put into the set E = {e i |i∈I neighbours }, e i is the insect species or frog species corresponding to the i-th sample; (315) According to the insect species or frog species in the set E, the insect species or frog species corresponding to the feature vector C is determined by majority voting to obtain the insect species information or frog species information: Where: is the indicator function, when c i When =c, the result is 1, otherwise it is 0.

6. The method for real-time monitoring and early warning of farmland insect pests according to claim 1, characterized in that: In the step (3), sampling the sound signal according to the insect species information and the frog species information, and determining the insect quantity information and the frog quantity information comprises the following steps: (321) According to the information of insect species or frog species, the average time T for the insect or frog to call once is taken, and the sound signal A(t) is collected at intervals T, and a total of M times is collected to obtain the discrete amplitude s i : s i =A(t i ),i∈[1,M] Where: A(t i ) is the sound signal collected for the i-th time; (322) The sound signal s i Converted into digital signal s′ i : s′ i =ADC(s i ) (323) According to the digital signal s′ i Calculate the energy E k : (324) Set the energy threshold. If the energy E k If it is greater than the energy threshold, it is determined as an event, and the total number of events in M ​​time periods T is calculated to obtain the corresponding insect quantity information or frog quantity information.

7. The method for real-time monitoring and early warning of farmland insect pests according to claim 1, characterized in that: The step (4) comprises: When the number of frogs in the area is less than the set value, the warning threshold is set to 1 / 4 of the insect population carrying capacity K. When the warning threshold is exceeded, a warning message is sent to the user; When the number of frogs in the area is greater than or equal to the set value, an early warning threshold is set according to the number of frogs, and the early warning threshold is greater than 1 / 4K and less than 1 / 2K.

8. The method for real-time monitoring and early warning of farmland insect pests according to claim 1, characterized in that: The step (4) also includes: calculating the insect density in different time periods of the target area according to the insect quantity information, obtaining the time period with the highest insect density by statistics, and sending warning information to the user.

9. A real-time monitoring and early warning device for farmland pests based on acoustic detection, characterized in that: It includes several monitoring units and a central data processing unit, wherein the monitoring unit includes a microphone array, an audio preprocessing unit, and a local data processing unit, wherein the microphone array includes several symmetrically distributed microphones, the microphone array, the audio preprocessing unit, and the local data processing unit are connected in sequence, and the local data processing unit is connected to the central data processing unit; Microphone array: used to collect sound signals in the target area and transmit them to the audio preprocessing unit; Audio preprocessing unit: used to perform denoising on the sound signal, extract the feature vector of the sound signal, and transmit it to the local data processing unit; Local data processing unit: used to compare the characteristic vector of the sound signal with the standard data set by using a classification algorithm to obtain the insect species information and frog species information corresponding to the sound signal; sample the sound signal according to the insect species information and frog species information, determine the insect quantity information and frog quantity information, and transmit them to the central data processing unit; Central data processing unit: used to set warning thresholds based on frog population information and insect population information, and to issue an alarm when the warning thresholds are exceeded.

10. The real-time monitoring and early warning device for farmland insect pests according to claim 9, characterized in that: The arrangement of the microphone array includes: uniform grid distribution: each microphone is arranged equidistantly in a rectangular grid; boundary reinforcement distribution: the number of microphones is increased at the boundary of the target area and the number of microphones is reduced at the center; annular distribution: the microphones are arranged in a circle or multiple concentric circles around the center of the target area.

Images