CN111797781A - Plant disease and insect pest identification system based on image identification and BP neural network - Google Patents
Plant disease and insect pest identification system based on image identification and BP neural network Download PDFInfo
- Publication number
- CN111797781A CN111797781A CN202010653571.0A CN202010653571A CN111797781A CN 111797781 A CN111797781 A CN 111797781A CN 202010653571 A CN202010653571 A CN 202010653571A CN 111797781 A CN111797781 A CN 111797781A
- Authority
- CN
- China
- Prior art keywords
- pest
- image
- disease
- neural network
- output
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
Abstract
The invention discloses a pest and disease identification system based on image identification and a BP neural network, which comprises a pest and disease image acquisition module, a pest and disease image processing module, a pest and disease image receiving and storing module, a pest and disease image identification and processing module and a feedback module. Has the advantages that: according to the invention, the BP neural network algorithm is adopted to identify and process the pest and disease damage image, so that in the specific use process, the pest and disease damage can be found more effectively and rapidly, the position where the pest and disease damage occurs can be positioned, and a farmer can spray pesticide to the area where the pest and disease damage occurs at the highest speed, thereby avoiding the situation that the pest and disease damage occurs in a large area, and simultaneously avoiding the situation that the income of the farmer is reduced sharply.
Description
Technical Field
The invention relates to the field of pest and disease identification systems, in particular to a pest and disease identification system based on image identification and a BP neural network.
Background
Diseases and pests are the combined name of diseases and pests, and often have adverse effects on agriculture, forestry, animal husbandry and the like. During the cultivation of medicinal plants, the plants are affected by the adverse environmental conditions of harmful organisms, the normal metabolism is disturbed, and a series of changes and damages occur from physiological functions to tissue structures, so that abnormal pathological changes such as withering, rot, spots and the like are presented on external forms, which are collectively called diseases. The causes of the disease of the medicinal plants comprise biological factors and non-biological factors. Diseases caused by invasion of biological factors such as fungi, bacteria, viruses, etc. into plants are infectious diseases called invasive diseases or parasitic diseases, diseases caused by influence or damage of physiological functions by abiotic factors such as drought, waterlogging, severe cold, nutrient imbalance, etc. are not infectious diseases called non-invasive diseases or physiological diseases.
In our daily life, foods such as vegetables, fruits and cereals are indispensable, but in the process of growing vegetables, fruits, cereal plants and the like, the foods can be damaged by different pests and diseases and can be eaten, and further after the pests and diseases damage the plants such as the vegetables, the fruits, the cereals and the like, the harvest of the plants such as the vegetables, the fruits, the cereals and the like can be greatly reduced, and the income of farmers is reduced sharply, so that a system capable of helping farmers identify and feed back the types of the pests and diseases is needed.
An effective solution to the problems in the related art has not been proposed yet.
Disclosure of Invention
Aiming at the problems in the related art, the invention provides a pest and disease identification system based on image identification and a BP neural network, so as to overcome the technical problems in the prior related art.
Therefore, the invention adopts the following specific technical scheme:
a pest and disease identification system based on image identification and a BP neural network comprises a pest and disease image acquisition module, a pest and disease image processing module, a pest and disease image receiving and storing module, a pest and disease image identification and processing module and a feedback module.
Further, the pest image acquisition module comprises a shooting terminal, a compression code and pest image preprocessing operation;
the shooting terminal comprises a mobile phone terminal, a camera terminal and a video camera terminal;
the compressed code is used for initializing the pest and disease image shot by the shooting terminal, setting a series of parameters and starting to code the pest and disease image.
Further, the pest and disease image preprocessing operation comprises pest and disease image uploading, pest and disease image storage, GPRS and GPS;
wherein, the GPRS is used for the transmission of remote pest and disease images;
wherein, GPS is used for fixing a position the area that the plant diseases and insect pests appear.
Further, the identification and processing module of the disease and insect pest image comprises the steps of preprocessing the disease and insect pest image and processing the disease and insect pest image by adopting a BP neural network algorithm;
the pretreatment of the pest and disease damage image comprises the following steps:
step 1, adjusting the resolution, definition and pixels of a received pest and disease damage image;
step 2, carrying out normalization operation on the images of the received diseases and insect pests with different sizes;
step 3, making the adjusted pest and disease damage image data into an LMDB format;
the method comprises the following steps of (1) processing a pest and disease damage image by adopting a BP neural network algorithm, wherein the preprocessing, the initial processing and the output processing are included;
the preprocessing is used for extracting the superfine of the image and is realized by an algorithm;
wherein the initial processing comprises initialization and initial output;
the output processing is used for carrying out feedback transmission when errors occur in the layer-by-layer transmission neurons, the neurons can carry out reverse transmission when the output result does not reach the expected output, and in the process of the reverse transmission, the network weight and the threshold are adjusted according to the errors so as to achieve a good identification prediction system.
Further, the pretreatment comprises the following steps:
step 1, preprocessing an observed pest image X;
step 2, extracting an independent component, searching a separation matrix L, extracting the independent component through S-LX, adding the mean value of the observation signal removed in the step 1 to the extracted independent component to obtain a signal S, and knowing by a central limit theorem that if a random variable consists of the sum of a plurality of mutually independent random variables, if each independent random variable of S has a limited mean value and difference, the following definition of probability distribution is satisfied;
definition 1: h (x) ═ f (x) lgf (x) dx
Wherein, H (x) is the information entropy of the random variable x with probability density f (x);
definition 2: j (x) h (bxgausian) -h (x)
Wherein, h (bxgausian) is the information entropy of the random variable having the same variance as the random variable x, J is the whitening matrix, and h (x) is the information entropy of the random variable x having a probability density of f (x);
from the above definition, it can be seen that the negative entropy of random variables with gaussian distribution is minimal and zero, thereby providing a simplified algorithm as follows:
J(x)∞[E{G(x)-E(xgaussian)}]2
wherein E (.) is an averaging operation, E is a cross-over matrix, J is a whitening matrix, and xgaussianIs a random variable, and x is information entropy;
wherein, the available number of G (.) is as follows;
G2(u)＝-exp(-u2/2)
and a is1Is constant and satisfies a 1 ≦ a12 or less, in general a is taken1The independent component process is a non-gaussian process that measures random variables, 1.
Further, the initialization comprises the steps of:
step 1, determining the number n of nodes of a network input layer, the number 1 of hidden layer nodes and the number m of output layer nodes;
step 2, determining weight W between neuron of input layer and hidden layer and neuron of hidden layer and output layerijAnd Wjk；
And 3, determining a hidden layer threshold value a and outputting a layer threshold value b.
Further, the initial output comprises the following steps:
step 1, calculating a formula of hidden layer output H as follows:
wherein f is a hidden layer excitation function, H is a hidden layer output function, and the expression form in the design is as follows:
step two, outputting a function H and a weight W according to the hidden layerjkAnd a threshold b, the formula for calculating the neural network prediction output O is as follows:
wherein, O is the prediction output of the neural network, and H is the hidden layer output function.
Further, the output processing includes the steps of:
step 1, calculating errors, wherein the formula for calculating the errors is as follows:
ek＝Yk-Ok
k＝1,2，...m
wherein O is the network prediction output, Y is the expected output, e is the prediction error value of the calculated neural network, and k is a constant;
step 2, updating the weight value, wherein the formula for updating the weight value is as follows:
i＝1,2，...n；j＝1,2，...l
wjk＝wjk+ηHjek
k＝1,2，...m；j＝1,2，...l
updating the network connection weight according to the network prediction error e;
wherein i is a constant, j is a constant, H is a hidden layer output function, e is a prediction error value of the neural network, and k is a constant;
and step 3, updating the threshold, wherein the formula of the threshold updating is as follows:
j＝1,2，...l
bk＝bk+ek
k＝1,2，...m
updating the network node threshold values a and b according to the network prediction error e;
wherein j is a constant, k is a constant, and e is a neural network prediction error.
And 4, judging whether the algorithm is finished or not, when the error is smaller than a set value, predicting that the output approaches the real output, and finishing the algorithm.
Further, the feedback module is used for feeding back the name of the found pest, the geographical position where the pest occurs and the time when the pest occurs.
The invention has the beneficial effects that:
1. compared with the traditional disease and insect pest identification system, the disease and insect pest identification system has the advantages that the BP neural network algorithm is adopted to identify and process the disease and insect pest image, so that the disease and insect pest can be found more effectively and rapidly in the specific using process, the position where the disease and insect pest occurs can be positioned, a farmer can spray pesticide to the area where the disease and insect pest occurs at the highest speed, the condition that the disease and insect pest occurs in a large area is further avoided, and meanwhile, the condition that the income of the farmer is reduced is also avoided.
2. Compared with the traditional plant disease and insect pest identification system, the plant disease and insect pest image is identified and processed by adopting the image identification and BP neural network algorithm, so that in a specific use process, the plant disease and insect pest can be effectively identified while the plant disease and insect pest are effectively and rapidly found, accurate plant disease and insect pest information is provided for farmers, the speed of solving the plant disease and insect pest is increased, and the condition of reducing the yield of the farmers is avoided.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
Fig. 1 is a flow chart of a pest and disease identification system based on image identification and a BP neural network according to an embodiment of the present invention;
FIG. 2 shows image acquisition of a pest and disease damage identification system based on image identification and BP neural network according to an embodiment of the present invention
FIG. 3 shows a pest and disease damage identification system based on image identification and BP neural network according to an embodiment of the invention
FIG. 4 shows a pest and disease damage identification system based on image identification and BP neural network according to an embodiment of the invention
FIG. 5 shows a pest and disease damage identification system based on image identification and BP neural network according to an embodiment of the invention
In the figure:
1. a pest image acquisition module; 101. shooting a terminal; 102. compression encoding; 103. preprocessing the pest and disease image; 104. a mobile phone terminal; 105. a camera terminal; 106. a camera terminal; 107. uploading a pest image; 108. storing the pest and disease damage image; 109. GPRS (general packet radio service); 110. a GPS; 2. a pest image processing module; 3. a pest and disease damage image receiving and storing module; 4. the pest and disease damage image identification and processing module; 401. preprocessing a pest and disease damage image; 402. processing the pest and disease damage image by adopting a BP neural network algorithm; 403. pre-treating; 404. carrying out initial treatment; 405. output processing; 406. initializing; 407. initial output; 5. and a feedback module.
Detailed Description
For further explanation of the various embodiments, the drawings which form a part of the disclosure and which are incorporated in and constitute a part of this specification, illustrate embodiments and, together with the description, serve to explain the principles of operation of the embodiments, and to enable others of ordinary skill in the art to understand the various embodiments and advantages of the invention, and, by reference to these figures, reference is made to the accompanying drawings, which are not to scale and wherein like reference numerals generally refer to like elements.
According to the embodiment of the invention, the pest and disease damage identification system based on the image identification and the BP neural network is provided.
Referring to the drawings and the detailed description, as shown in fig. 1 to 5, the pest and disease identification system based on image identification and BP neural network according to the embodiment of the present invention includes a pest and disease image acquisition module 1, a pest and disease image processing module 2, a pest and disease image receiving and storing module 3, a pest and disease image identification and processing module 4, and a feedback module 5.
In one example, the pest image acquisition module 1 comprises a shooting terminal 101, a compression code 102 and a pest image preprocessing operation 103;
the shooting terminal 101 comprises a mobile phone terminal 104, a camera terminal 105 and a video camera terminal 106;
the compression code 102 is used for initializing the pest image shot by the shooting terminal 101, setting a series of parameters, and starting to code the pest image.
In one example, the pest image preprocessing operation 103 includes pest image uploading 107, pest image saving 108, GPRS109, and GPS 110;
wherein, the GPRS109 is used for transmitting remote pest and disease images;
wherein, GPS110 is used for fixing a position the area that the pest appears.
In one example, the pest and disease damage image identification and processing module 4 comprises a pest and disease damage image preprocessing module 401 and a pest and disease damage image processing module 402 which adopts a BP neural network algorithm;
the pretreatment 401 of the pest and disease damage image comprises the following steps:
step 1, adjusting the resolution, definition and pixels of a received pest and disease damage image;
step 2, carrying out normalization operation on the images of the received diseases and insect pests with different sizes;
step 3, making the adjusted pest and disease damage image data into an LMDB format;
the method comprises the following steps that (1) a BP neural network algorithm is adopted to process 402 the pest and disease damage image, wherein the preprocessing 403, the initial processing 404 and the output processing 405 are included;
the preprocessing 403 is used for extracting the minuteness of the image and is implemented by using an algorithm;
wherein the initial processing 404 includes initialization 406 and initial output 407;
the output processing 405 is configured to perform feedback transmission when an error occurs in the layer-by-layer transmission neurons, and when an output result does not reach an expected output, the neurons may perform reverse transmission, and in the process of reverse transmission, a network weight and a threshold are adjusted according to the error, so as to achieve a good recognition prediction system.
In one example, the pre-processing 403 includes the steps of:
step 1, preprocessing an observed pest image X;
step 2, extracting an independent component, searching a separation matrix L, extracting the independent component through S-LX, adding the mean value of the observation signal removed in the step 1 to the extracted independent component to obtain a signal S, and knowing by a central limit theorem that if a random variable consists of the sum of a plurality of mutually independent random variables, if each independent random variable of S has a limited mean value and difference, the following definition of probability distribution is satisfied;
definition 1: h (x) ═ f (x) lgf (x) dx
Wherein, H (x) is the information entropy of the random variable x with probability density f (x);
definition 2: j (x) h (bxgausian) -h (x)
Wherein, h (bxgausian) is the information entropy of the random variable having the same variance as the random variable x, J is the whitening matrix, and h (x) is the information entropy of the random variable x having a probability density of f (x);
from the above definition, it can be seen that the negative entropy of random variables with gaussian distribution is minimal and zero, thereby providing a simplified algorithm as follows:
J(x)∞[E{G(x)-E(xgaussian)}]2
wherein E (.) is an averaging operation, E is a cross-over matrix, J is a whitening matrix, and xgaussianIs a random variable, and x is information entropy;
wherein, the available number of G (.) is as follows;
G2(u)＝-exp(-u2/2)
and a is1Is constant and satisfies a 1 ≦ a12 or less, in general a is taken1The independent component process is a non-gaussian process that measures random variables, 1.
In one example, the initialization 406 includes the steps of:
step 1, determining the number n of nodes of a network input layer, the number 1 of hidden layer nodes and the number m of output layer nodes;
step 2, determining weight W between neuron of input layer and hidden layer and neuron of hidden layer and output layerijAnd Wjk；
And 3, determining a hidden layer threshold value a and outputting a layer threshold value b.
In one example, the initial output 407 includes the following steps:
step 1, calculating a formula of hidden layer output H as follows:
wherein f is a hidden layer excitation function, H is a hidden layer output function, and the expression form in the design is as follows:
step two, outputting a function H and a weight W according to the hidden layerjkAnd a threshold b, the formula for calculating the neural network prediction output O is as follows:
wherein, O is the prediction output of the neural network, and H is the hidden layer output function.
In one example, the output process 405 includes the steps of:
step 1, calculating errors, wherein the formula for calculating the errors is as follows:
ek＝Yk-Ok
k＝1,2，...m
wherein O is the network prediction output, Y is the expected output, e is the prediction error value of the calculated neural network, and k is a constant;
step 2, updating the weight value, wherein the formula for updating the weight value is as follows:
i＝1,2，...n；j＝1,2，...l
wjk＝wjk+ηHjek
k＝1,2，...m；j＝1,2，...l
updating the network connection weight according to the network prediction error e;
wherein i is a constant, j is a constant, H is a hidden layer output function, e is a prediction error value of the neural network, and k is a constant;
and step 3, updating the threshold, wherein the formula of the threshold updating is as follows:
j＝1,2，...l
bk＝bk+ek
k＝1,2，...m
updating the network node threshold values a and b according to the network prediction error e;
wherein j is a constant, k is a constant, and e is a neural network prediction error.
And 4, judging whether the algorithm is finished or not, when the error is smaller than a set value, predicting that the output approaches the real output, and finishing the algorithm.
In one example, the feedback module 5 is used to feedback the name of the pest found, the geographical location where the pest occurred, and the time at which the pest occurred.
For the convenience of understanding the above technical solutions of the present invention, the following detailed descriptions of the above solutions of the present invention are provided in conjunction with examples, and specifically, the following are provided:
1 example one
1.1 selection of objects
The designed system selects the gray leaf spot of the soybean to test.
By comparing the understanding of the soybean gray leaf spot with the normal soybeans, the color and the texture characteristics of the soybeans with the soybean gray leaf spot are different to a certain extent, and the difference between the color and the texture characteristics of the soybeans with the soybean gray leaf spot is used as a classification basis.
1.2 extraction of features
1.2.1 extraction of color features
Selecting an RGB color system, wherein the RGB colors are based on a model formed by human vision, and all colors are formed by: in soybean diseases and insect pests, color characteristics of different diseases are mostly distributed in a low-order matrix. Therefore, the first moment sigma of an RGB color system capable of extracting the pest and disease damage image is adopted1And second order moment sigma2：
In the above formula Ii(x, y) represents the gray scale of the image.
The first moment and the second moment of RGB three-channel components are extracted from soybean pest and disease images and normal images by using MARLAB software, and the first moment and the second moment are shown in table 1.
TABLE 1 image color matrix characterization parameters
It can be seen from table 1 that the first moment of the soybean image of the normal soybean is slightly lower than that of the soybean image of the soybean gray leaf spot, but the R component and the B component have a significant difference; in the second moment, the normal soybean moment is significantly lower than that of soybean with gray leaf spot of soybean, and there is mainly a G component.
1.2.2 extraction of textural features
The characteristic reflected by the texture is locally slowly changed or repeated, and based on the characteristic, the texture characteristic of different images can be extracted for identification. Texture features are represented by pixels and local texture information, i.e. the grey scale distribution of the surrounding space. The design adopts a gray level co-occurrence matrix method to extract textures.
TABLE 2 texture feature parameters
From table 2, the entropy values of normal soybeans are tens of times higher than the soybean images of soybean gray spot disease, but at this time, the standard deviation of the images of soybeans of soybean gray spot disease is large, and the normal soybeans are always higher than the large images of soybeans of soybean gray spot disease in terms of energy, moment of inertia or correlation.
2 example one
2.1 selection of objects
The designed system selects the corn rust spot to test.
By comparing the understanding of the rust spot with the normal corn, the color and the texture characteristics of the corn with the rust spot and the normal corn with certain differences can be found, and the difference between the color and the texture characteristics of the corn with the rust spot and the normal corn can be used as the basis for classification.
2.2 extraction of features
2.2.1 extraction of color features
Selecting an RGB color system, wherein the RGB colors are based on a model formed by human vision, and all colors are formed by: in soybean diseases and insect pests, color characteristics of different diseases are mostly distributed in a low-order matrix. Therefore, the first moment and the second moment of an RGB color system of the disease and insect pest image can be extracted:
in the above formula Ii(x, y) represents the gray scale of the image.
The first moment and the second moment of RGB three-channel components are extracted from the corn pest and disease damage image and the normal image by using MARLAB software, and a table 3 is formed.
TABLE 3 image color matrix characterization parameters
It can be seen from table 3 that the normal corn has a slightly lower first moment than the corn image of russet leaf spot, but there is also a significant difference between the R and B components; in the second moment, the normal second moment is significantly lower than that of rustic spot, and there is mainly a G component.
2.2.2 extraction of textural features
The characteristic reflected by the texture is locally slowly changed or repeated, and based on the characteristic, the texture characteristic of different images can be extracted for identification. Texture features are represented by pixels and local texture information, i.e. the grey scale distribution of the surrounding space. The design adopts a gray level co-occurrence matrix method to extract textures.
FIG. 4 texture feature parameters
From table 4, normal corn is tens of times higher in entropy than the corn image of rusty spot, but at this time, the standard deviation of the image of rusty spot is large, and normal corn is always higher than the large image of rusty spot soybean, regardless of energy, moment of inertia or correlation.
3 conclusion
The invention can effectively carry out image preprocessing and timely feedback, and the identification accuracy of the invention is up to more than eighty percent.
In summary, by means of the technical scheme, compared with the traditional disease and insect pest identification system, the disease and insect pest identification system has the advantages that the BP neural network algorithm is adopted to identify and process the disease and insect pest image, so that in the specific use process, the disease and insect pest can be found more effectively and rapidly, the position where the disease and insect pest occurs can be located, a farmer can spray the pesticide to the area where the disease and insect pest occurs at the fastest speed, the condition that the disease and insect pest occurs in a large area is avoided, and meanwhile, the condition that the income of the farmer is reduced sharply is avoided. Compared with the traditional plant disease and insect pest identification system, the plant disease and insect pest image is identified and processed by adopting the image identification and BP neural network algorithm, so that in a specific use process, the plant disease and insect pest can be effectively identified while the plant disease and insect pest are effectively and rapidly found, accurate plant disease and insect pest information is provided for farmers, the speed of solving the plant disease and insect pest is increased, and the condition of reducing the yield of the farmers is avoided.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (9)
1. The pest and disease identification system based on the image identification and BP neural network is characterized by comprising a pest and disease image acquisition module (1), a pest and disease image processing module (2), a pest and disease image receiving and storing module (3), a pest and disease image identification and processing module (4) and a feedback module (5).
2. A pest and disease identification system based on image identification and BP neural network according to claim 1, wherein the pest and disease image acquisition module (1) comprises a shooting terminal (101), a compression coding (102) and a pest and disease image preprocessing operation (103);
the shooting terminal (101) comprises a mobile phone terminal (104), a camera terminal (105) and a video camera terminal (106);
the compression coding (102) is used for initializing the pest and disease image shot by the shooting terminal (101), setting a series of parameters and starting to code the pest and disease image.
3. A pest identification system based on image identification and BP neural network according to claim 2, wherein the pest image preprocessing operation (103) comprises pest image uploading (107), pest image saving (108), GPRS (109) and GPS (110);
wherein the GPRS (109) is used for long-distance transmission of pest and disease images;
wherein, the GPS (110) is used for positioning the pest occurrence area.
4. A pest identification system based on image identification and BP neural network according to claim 1, wherein the pest image identification and processing module (4) comprises pest image preprocessing (401) and pest image processing (402) by using BP neural network algorithm;
wherein the preprocessing (401) of the pest image comprises the following steps:
step 1, adjusting the resolution, definition and pixels of a received pest and disease damage image;
step 2, carrying out normalization operation on the images of the received diseases and insect pests with different sizes;
step 3, making the adjusted pest and disease damage image data into an LMDB format;
processing (402) the pest and disease damage image by adopting a BP neural network algorithm comprises preprocessing (403), initial processing (404) and output processing (405);
wherein the preprocessing (403) is used for extracting the superfine of the image and is realized by an algorithm;
wherein the initial processing (404) comprises an initialization (406) and an initial output (407);
the output processing (405) is used for performing feedback transmission when errors occur in the layer-by-layer transmission neurons, the neurons can perform reverse transmission when the output result does not reach the expected output, and in the process of the reverse transmission, the network weight and the threshold are adjusted according to the errors so as to achieve a good recognition prediction system.
5. An image recognition and BP neural network based pest identification system according to claim 4, wherein the pre-processing (403) comprises the steps of:
step 1, preprocessing an observed pest image X;
step 2, extracting an independent component, searching a separation matrix L, extracting the independent component through S-LX, adding the mean value of the observation signal removed in the step 1 to the extracted independent component to obtain a signal S, and knowing by a central limit theorem that if a random variable consists of the sum of a plurality of mutually independent random variables, if each independent random variable of S has a limited mean value and difference, the following definition of probability distribution is satisfied;
definition 1: h (x) ═ f (x) lgf (x) dx
Wherein, H (x) is the information entropy of the random variable x with probability density f (x);
definition 2: j (x) h (bxgausian) -h (x)
Wherein, h (bxgausian) is the information entropy of the random variable having the same variance as the random variable x, J is the whitening matrix, and h (x) is the information entropy of the random variable x having a probability density of f (x);
from the above definition, it can be seen that the negative entropy of random variables with gaussian distribution is minimal and zero, thereby providing a simplified algorithm as follows:
J(x)∞[E{G(x)-E(xgaussian)}]2
wherein E (.) is an averaging operation, E is a cross-over matrix, J is a whitening matrix, and xgaussianIs a random variable, and x is information entropy;
wherein, the available number of G (.) is as follows;
G2(u)＝-exp(-u2/2)
and a is1Is constant and satisfies a 1 ≦ a12 or less, in general a is taken1The independent component process is a non-gaussian process that measures random variables, 1.
6. An image recognition and BP neural network based pest identification system according to claim 4, wherein the initialization (406) comprises the steps of:
step 1, determining the number n of nodes of a network input layer, the number 1 of hidden layer nodes and the number m of output layer nodes;
step 2, determining weight W between neuron of input layer and hidden layer and neuron of hidden layer and output layerijAnd Wjk；
And 3, determining a hidden layer threshold value a and outputting a layer threshold value b.
7. An image recognition and BP neural network based pest identification system according to claim 4, wherein the initial output (407) comprises the steps of:
step 1, calculating a formula of hidden layer output H as follows:
wherein f is a hidden layer excitation function, H is a hidden layer output function, and the expression form in the design is as follows:
step two, outputting a function H and a weight W according to the hidden layerjkAnd a threshold b, the formula for calculating the neural network prediction output O is as follows:
wherein, O is the prediction output of the neural network, and H is the hidden layer output function.
8. An image recognition and BP neural network based pest identification system according to claim 4, wherein the output process (405) comprises the steps of:
step 1, calculating errors, wherein the formula for calculating the errors is as follows:
ek＝Yk-Ok
k＝1,2，...m
wherein O is the network prediction output, Y is the expected output, e is the prediction error value of the calculated neural network, and k is a constant;
step 2, updating the weight value, wherein the formula for updating the weight value is as follows:
i＝1,2，...n；j＝1,2，...l
wjk＝wjk+ηHjek
k＝1,2，...m；j＝1,2，...l
updating the network connection weight according to the network prediction error e;
wherein i is a constant, j is a constant, H is a hidden layer output function, e is a prediction error value of the neural network, and k is a constant;
and step 3, updating the threshold, wherein the formula of the threshold updating is as follows:
j＝1,2，...l
bk＝bk+ek
k＝1,2，...m
updating the network node threshold values a and b according to the network prediction error e;
wherein j is a constant, k is a constant, and e is a neural network prediction error.
And 4, judging whether the algorithm is finished or not, when the error is smaller than a set value, predicting that the output approaches the real output, and finishing the algorithm.
9. A pest identification system based on image identification and BP neural network according to claim 1, wherein said feedback module (5) is used for feeding back the name of the pest, the geographical location where the pest occurs and the time when the pest occurs.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010653571.0A CN111797781A (en) | 2020-07-08 | 2020-07-08 | Plant disease and insect pest identification system based on image identification and BP neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010653571.0A CN111797781A (en) | 2020-07-08 | 2020-07-08 | Plant disease and insect pest identification system based on image identification and BP neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111797781A true CN111797781A (en) | 2020-10-20 |
Family
ID=72810600
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010653571.0A Pending CN111797781A (en) | 2020-07-08 | 2020-07-08 | Plant disease and insect pest identification system based on image identification and BP neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111797781A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114397929A (en) * | 2022-01-18 | 2022-04-26 | 中山东菱威力电器有限公司 | Intelligent toilet lid control system capable of improving initial temperature of flushing water |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105320970A (en) * | 2015-11-25 | 2016-02-10 | 内蒙古科技大学 | Potato disease diagnostic device, diagnostic system and diagnostic method |
| CN105844285A (en) * | 2016-02-25 | 2016-08-10 | 中国农业大学 | Cucumber disease identification method and apparatus based on image information |
| CN108921849A (en) * | 2018-09-30 | 2018-11-30 | 靖西海越农业有限公司 | For preventing and treating the wisdom Agricultural Monitoring early warning system of fertile mandarin orange pest and disease damage |
-
2020
- 2020-07-08 CN CN202010653571.0A patent/CN111797781A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105320970A (en) * | 2015-11-25 | 2016-02-10 | 内蒙古科技大学 | Potato disease diagnostic device, diagnostic system and diagnostic method |
| CN105844285A (en) * | 2016-02-25 | 2016-08-10 | 中国农业大学 | Cucumber disease identification method and apparatus based on image information |
| CN108921849A (en) * | 2018-09-30 | 2018-11-30 | 靖西海越农业有限公司 | For preventing and treating the wisdom Agricultural Monitoring early warning system of fertile mandarin orange pest and disease damage |
Non-Patent Citations (2)
| Title |
|---|
| 邓祺盛 等: "基于BP神经网络的图像识别系统", 《现代计算机》 * |
| 靳洋: "基于PCA/ICA的图像特征提取算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114397929A (en) * | 2022-01-18 | 2022-04-26 | 中山东菱威力电器有限公司 | Intelligent toilet lid control system capable of improving initial temperature of flushing water |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| AU2020103613A4 (en) | Cnn and transfer learning based disease intelligent identification method and system | |
| Sabzi et al. | An automatic visible-range video weed detection, segmentation and classification prototype in potato field | |
| Flores et al. | Distinguishing seedling volunteer corn from soybean through greenhouse color, color-infrared, and fused images using machine and deep learning | |
| EP3942465B1 (en) | Industrialized system for rice grain recognition and method thereof | |
| CN111860330B (en) | Apple leaf disease identification method based on multi-feature fusion and convolutional neural network | |
| CN105989601B (en) | Agricultural AGV corn inter-row navigation datum line extraction method based on machine vision | |
| Gulavnai et al. | Deep learning for image based mango leaf disease detection | |
| Arshad et al. | Plant disease identification using transfer learning | |
| Sheikh et al. | Detection of maize and peach leaf diseases using image processing | |
| CN113627282A (en) | Tea disease identification method based on deep migration learning | |
| CN111797781A (en) | Plant disease and insect pest identification system based on image identification and BP neural network | |
| Patil | Pomegranate fruit diseases detection using image processing techniques: a review | |
| CN112465038A (en) | Method and system for identifying disease and insect pest types of fruit trees | |
| CN115511158B (en) | Intelligent crop breeding analysis method and system based on big data | |
| Kumar et al. | An automatic multi-threshold image processing technique for mushroom disease segmentation | |
| Juyal et al. | Strawberry plant's health detection for organic farming using unmanned aerial vehicle | |
| Deng et al. | Estimating the severity of sugarcane aphids infestation on sorghum with machine vision | |
| CN113298835B (en) | Fruit harvesting maturity weighted evaluation method | |
| Mladenov et al. | Quality assessment of grain samples using color image analysis | |
| CN114462485A (en) | Red date jujube witches broom initial-stage control method | |
| Rodriguez et al. | Machine learning techniques in the detection of cocoa (Theobroma cacao L.) diseases | |
| Kaesmetan et al. | Digital Image Processing using Texture Features Extraction of Local Seeds in Nekbaun Village with Color Moment, Gray Level Co Occurance Matrix, and k-Nearest Neighbor | |
| CN111986149A (en) | Plant disease and insect pest detection method based on convolutional neural network | |
| CN114078228A (en) | Target identification method and system based on lightweight network and agricultural machine | |
| CN112580610A (en) | Banana wilt remote sensing rapid detection method based on full convolution neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |