CN-122020554-A - Automatic identification method and model for tobacco diseases

CN122020554ACN 122020554 ACN122020554 ACN 122020554ACN-122020554-A

Abstract

The invention discloses an automatic identification method and model for tobacco diseases, comprising the following steps of calculating the probability of infectious diseases K based on image features, acquiring non-image features, calculating non-image feature adjustment factors, calculating comprehensive adjustment factors, calculating the probability of occurrence, arranging from large to small according to the probability of occurrence of the diseases, outputting the probability of the first N diseases and the disease types, and displaying corresponding image features and non-image features. The method has the advantages that the limitation of single image analysis is broken through, and a multi-source information fusion decision framework of image characteristics, environmental climate, growth stage and history condition is constructed.

Inventors

XIA TIYUAN
FAN ZHIWEI
WANG YANHONG
ZHONG YU
TANG ZUOXIN
WU CHENG
PAN YIHONG
ZHU XIAOHONG
Ren Die
ZHANG YU

Assignees

昆明学院

Dates

Publication Date: 20260512
Application Date: 20260204

Claims (10)

1. The automatic identification method for the tobacco diseases is characterized by comprising the following steps: S1, calculating the probability of the infection disease K based on image features, namely transmitting the acquired crop image to a trained deep learning diagnosis model, wherein the model directly outputs the preliminary probability distribution of symptoms presented by the image belonging to various diseases, and the preliminary probability distribution is marked as P k (image);P k (image) and represents the probability of the infection disease K of the crop, which is obtained based on image feature judgment; s2, acquiring non-image characteristics, wherein the non-image characteristics comprise weather data of the past N days, current growth stages and actual incidence rate of the historical diseases for nearly N years, the weather data comprise temperature, humidity, illumination, wind speed and precipitation of each day, the growth stages are days after transplanting, and the transplanting day is the first day; S3, calculating a non-image characteristic adjustment factor, namely calculating the non-image characteristic adjustment factor for the disease K according to the non-image characteristic, wherein the non-image characteristic adjustment factor comprises a climate matching degree C k , a growth stage matching degree G k (t) and a historical disease weight H k ; S4, calculating a comprehensive adjustment factor T k : T k = β 1k *C k + β 2k *G k (t) + β 3k *H k , and β 1k +β 2k +β 3k =1; Wherein beta 1k 、β 2k 、β 3k is the weight of the climate, growth stage and history disease transfer of the disease K respectively, and beta 1k 、β 2k 、β 3k is more than or equal to 0; S5, calculating the incidence probability P K : ; S6, arranging the diseases from large to small according to the incidence probability P K of the diseases, outputting the probability of the first N diseases and the disease types, and displaying the corresponding image characteristics and non-image characteristics.
2. The method for automatically identifying tobacco diseases according to claim 1, wherein the climate matching degree C k in the step S3 is calculated as follows: S31, calculating temperature matching degree Mtemp, humidity matching degree Mrh, illumination matching degree Mlight, wind speed matching degree Mwind and precipitation matching degree Mrain respectively; s32, determining the relative weights of temperature, humidity, illumination, wind speed and precipitation, wherein the relative weights are respectively represented by alpha temp, alpha rh, alpha light, alpha wind and alpha rain, and alpha temp+alpha rh+alpha light+alpha wind+alpha rain=1; s33, calculating C k ,C k as a weighted geometric mean of five climate factors including temperature, humidity, illumination, wind speed and precipitation; C k = (Mtemp^αtemp) × (Mrh^αrh) × (Mlight^αlight) × (Mwind^αwind) × (Mrain^αrain)。
3. The method for automatically identifying tobacco diseases according to claim 1, wherein the step S3 of calculating the growth stage matching degree Gk (t) is as follows: ; Wherein: qk represents the basic matching degree of disease k, and Qk is more than or equal to 0 and less than or equal to 0.3; rk represents the peak susceptibility of disease k, and represents the theoretical maximum matching degree in the optimal period, wherein Rk is more than or equal to 0 and less than or equal to 1; t 0,k represents the center point of the most susceptible period of disease k, which means that disease k is most susceptible to occurrence in days t 0,k after transplanting; ρk represents a time width parameter of disease k, controlling the rate of susceptibility decay with time in days, the larger the value, the longer the disease susceptibility duration; the values of Qk, rk, t 0,k and ρk for different diseases K are set in advance.
4. The method for automatically identifying tobacco diseases according to claim 1, wherein the method for calculating the historical disease weight Hk in the step S3 is as follows: ; Wherein: n represents the number of years of history considered; i represents a historical year number, i=1 represents the last year, and i=n represents the last year; lambda represents an attenuation factor, and controls the attenuation speed of the history influence, wherein lambda is more than 0 and less than or equal to 1, and the history influence is more durable when lambda is larger; I k,i represents the standardized incidence rate of disease k in the I-th year, I k,i is more than or equal to 0 and less than or equal to 1; I k,i =P k,i /P k,max ; P k,i represents the actual incidence of disease k in the ith year; P k,max represents the theoretical maximum incidence of disease k.
5. The method for automatically identifying tobacco diseases according to claim 1, wherein the method for determining beta 1k 、β 2k 、β 3k in the step S4 is to manually establish a weight matrix for different disease types.
6. The method for automatically identifying tobacco diseases according to claim 1, wherein the specific method in the step S1 is as follows: S11, collecting original images of field flue-cured tobacco, wherein the original images comprise an integral image of a plant, a stalk image and a plurality of leaf images of the same plant; s12, preprocessing an image, namely preprocessing an original image of the flue-cured tobacco in the acquired field; s13, calculating P leaf-n(k) , namely respectively extracting blade characteristics for each blade image, and calculating the incidence probability of each blade for different diseases K, namely P leaf-n(k) , wherein K represents the disease and leaf-n represents the nth blade of the plant; S14, aggregating leaf evidence, namely calculating to obtain weighted average incidence probability P leaf(k) of the leaves according to P leaf-n(k) of a plurality of leaves; S15, calculating P stem(k) , namely extracting stalk characteristics according to stalk images of plants, and calculating to obtain stalk incidence probability P stem(k) of the plants aiming at different diseases K; S16, calculating P whole(k) , namely extracting integral features according to integral images of plants, and calculating to obtain integral plant incidence probability P whole(k) of the plants aiming at different diseases K; s17, according to the calculated P leaf(k) 、P stem(k) 、P whole(k) , calculating to obtain the probability P k (image) of the plant against the disease K.
7. The method according to claim 1, wherein the leaf features include at least one of area ratio, distribution dispersion, weighted average circularity, weighted average elongation, weighted average fractal dimension, hue deviation, and chromaticity dispersion of the leaf, the plant features include at least one of canopy structure volume index, yellowing ratio of leaf, and top tip wilting index, and the stalk features include at least one of stalk base blackening ratio, stalk longitudinal streak index, and stalk base diameter anomaly rate.
8. The method for automatically identifying tobacco diseases according to claim 1, wherein the calculation method of P leaf-(k) in the step S14 comprises the following steps: s141, calculating weights Wn of a plurality of blades: , Alesion, n is the area of the nth blade; Aleaf, n is the total area of a plurality of lesions on the nth leaf; Is a constant; S142, calculating P leaf(k) : ; Wherein: n is the total number of leaves analyzed; P leaf-n(k) is the incidence probability of the nth leaf on the disease K; wn is the weight of the nth leaf.
9. The method for automatically identifying tobacco diseases according to claim 1, wherein the calculation method of P k (image) in the step S17 comprises the following steps: s171, determining weights of the leaf, the stem and the whole body, which are respectively marked as w leaf(k) ,w stem(k) ,w whole(k) , and the method is as follows: S1711, presetting an initial weight matrix of a leaf, a stem and a whole plant for different diseases k according to experience by an expert, wherein the initial weight matrix is denoted as w leaf(k)1 ,w stem(k)1 ,w whole(k)1 and satisfies the condition that w leaf(k)1 +w stem(k)1 +w whole(k)1 =1; S1712, respectively presetting a threshold range of incidence probability for the leaves, stems and whole plants of each disease k; S1713, in the actual diagnosis, comparing the calculated P leaf(k) 、P stem(k) 、P whole(k) with the respective incidence probability threshold range respectively, and if the incidence probability threshold range is exceeded, adjusting an initial weight matrix to obtain a final weight matrix, wherein the adjustment rule is that if the probability value is higher than the upper limit, the symptom of the part is very typical, the weight is increased, and if the probability value is lower than the lower limit, the symptom of the part is not obvious, and the weight is reduced; S1714, normalizing the adjusted weights to obtain a final weight matrix w leaf(k) ,w stem(k) ,w whole(k) of the leaves, the stems and the whole plant; s172, calculate P k (image): 。
10. the automatic tobacco disease identification model is characterized in that the automatic tobacco disease identification method of any one of claims 1-9 can be realized.

Description

Automatic identification method and model for tobacco diseases Technical Field The invention relates to the technical field of automatic identification of diseases, in particular to an automatic identification method and model of tobacco diseases. Background Currently, automatic identification of tobacco diseases mainly depends on image processing and deep learning technologies, and classification is performed by analyzing visual characteristics such as leaf color, texture, shape and the like. However, such methods have significant limitations in that, on the one hand, many diseases are highly similar in visual symptoms (e.g., brown spot and wildfire spot) and are extremely prone to misjudgement by images alone, and on the other hand, it ignores the agroecological nature of disease occurrence-disease prevalence is a result of pathogen, host and environmental interactions, which occur in close association with specific climatic conditions, tobacco growth stages and field history conditions. By means of image recognition alone, when early symptoms of diseases are not obvious, symptoms are similar or image quality is poor, misjudgment risk is high and early warning is delayed. Therefore, the reliability and early warning capability of the system are improved, the limitation of single image analysis must be broken through, and a multi-source intelligent decision model integrating real-time image characteristics, environmental climate data, crop growth period and historical disease information is constructed, so that more accurate and more robust automatic diagnosis is realized. Disclosure of Invention The invention aims to provide a tobacco disease automatic identification method and model which are integrated with real-time image characteristics, environmental climate data, crop growth period and historical disease information. In order to solve the technical problems, the technical scheme of the invention is an automatic tobacco disease identification method, which comprises the following steps: S1, calculating the probability of the infection disease K based on image features, namely transmitting the acquired crop image to a trained deep learning diagnosis model, wherein the model directly outputs the preliminary probability distribution of symptoms presented by the image belonging to various diseases, and the preliminary probability distribution is marked as P k(image);Pk (image) and represents the probability of the infection disease K of the crop, which is obtained based on image feature judgment; s2, acquiring non-image characteristics, wherein the non-image characteristics comprise weather data of the past N days, current growth stages and actual incidence rate of the historical diseases for nearly N years, the weather data comprise temperature, humidity, illumination, wind speed and precipitation of each day, the growth stages are days after transplanting, and the transplanting day is the first day; S3, calculating a non-image characteristic adjustment factor, namely calculating the non-image characteristic adjustment factor for the disease K according to the non-image characteristic, wherein the non-image characteristic adjustment factor comprises a climate matching degree C k, a growth stage matching degree G k (t) and a historical disease weight H k; S4, calculating a comprehensive adjustment factor T k: T k = β1k*Ck + β2k*Gk(t) + β3k*Hk, and β 1k+β2k+β3k =1; Wherein beta 1k、β2k、β3k is the weight of the climate, growth stage and history disease transfer of the disease K respectively, and beta 1k、β2k、β3k is more than or equal to 0; S5, calculating the incidence probability P K: ; S6, arranging the diseases from large to small according to the incidence probability P K of the diseases, outputting the probability of the first N diseases and the disease types, and displaying the corresponding image characteristics and non-image characteristics. Further, the calculation process of the climate matching degree C k in the step S3 is as follows: S31, calculating temperature matching degree Mtemp, humidity matching degree Mrh, illumination matching degree Mlight, wind speed matching degree Mwind and precipitation matching degree Mrain respectively; s32, determining the relative weights of temperature, humidity, illumination, wind speed and precipitation, wherein the relative weights are respectively represented by alpha temp, alpha rh, alpha light, alpha wind and alpha rain, and alpha temp+alpha rh+alpha light+alpha wind+alpha rain=1; s33, calculating C k,Ck as a weighted geometric mean of five climate factors including temperature, humidity, illumination, wind speed and precipitation; Ck= (Mtemp^αtemp) × (Mrh^αrh) × (Mlight^αlight) × (Mwind^αwind) × (Mrain^αrain)。 Further, the method for calculating the growth stage matching degree Gk (t) in the step S3 is as follows: ; Wherein: qk represents the basic matching degree of disease k, and Qk is more than or equal to 0 and less than or equal to 0.3; rk represent