A potato planting disease and pest identification and analysis method based on image processing

Abstract

The present application relates to the field of agricultural planting pest identification technology, and relates to a potato planting pest identification and analysis method based on image processing. The present application adopts unmanned aerial vehicle grid partition cruise to shoot panoramic images, combines with the benchmark image of the planting record of the same period to compare pixels, screens abnormal partitions and calibrates spatial positions, controls the unmanned aerial vehicle to collect the current canopy image corresponding to the abnormal partition at a close distance, compares the feature difference values of each leaf in the current canopy image with those in the historical canopy image, carries out similarity matching combined with the standard image feature difference matrix, judges the plant pest type of the abnormal partition, extracts the epidermal lesion morphological features and the cross-sectional decay area distribution features when the pest type cannot be determined, determines the final pest type of the abnormal partition, thereby realizing comprehensive disease evaluation of the whole organ of the potato plant, ensuring the comprehensiveness and integrity of the pest identification, and reducing the yield loss risk caused by the missed detection of tuber diseases.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of agricultural planting pest identification technology, and relates to a potato planting pest identification and analysis method based on image processing. BACKGROUND

[0002] Pests and diseases are one of the main risk factors affecting potato planting. Different types of pests and diseases show different development characteristics in the growth cycle of potatoes. If they cannot be identified in time, they will spread rapidly, causing plant growth to be hindered, tubers to rot and deteriorate, and causing serious yield loss. With the popularization of large-scale planting mode, intelligent pest and disease monitoring means based on image recognition has become a key technical direction to improve the management level of potato planting.

[0003] The prior art such as Chinese patent publication No. CN120655580A discloses a potato leaf disease area positioning method, system and device. The target infrared image of the potato leaf is segmented into a normal area and an abnormal area according to the environmental temperature through a pre-acquired disease segmentation model. The water content of the target infrared image is determined through a water content prediction model, and the emissivity characteristics of the leaf are calculated based on the water content through a temperature correction model to correct the temperature of the abnormal area. Finally, the disease area is identified from the abnormal area after temperature correction through a disease identification model.

[0004] However, the prior art has the following problems: 1. The prior art only positions the disease area based on the infrared image of the potato leaf, and does not involve the collection and analysis of tuber-related features, which has the problem of single identification dimension. Since the symptoms of some potato pests and diseases are manifested on both the leaf and the tuber, relying only on the leaf image cannot accurately identify the type of pests and diseases, which may lead to misjudgment of the disease type, affect the pertinence of subsequent prevention and control measures, and further cause poor prevention and control effect and the risk of pest and disease spread.

[0005] 2. The prior art uses pre-set machine learning models such as disease segmentation model and water content prediction model to achieve identification, and does not compare the time sequence characteristics in combination with the historical canopy image data of potato planting, which has the problem of lacking dynamic tracking of the development process of pests and diseases. The occurrence and development of pests and diseases is a time sequence process. Based on only the infrared image features at a single time point, the characteristic change rule of the disease evolution cannot be captured, which leads to insufficient identification sensitivity of early or occult pests and diseases, delays the best prevention and control opportunity, and causes irreversible loss of potato yield and quality. SUMMARY

[0006] The present application discloses a potato planting pest identification and analysis method based on image processing, which aims to solve the problems of single identification dimension and lack of dynamic tracking in the prior art.

[0007] The technical scheme adopted by the present application to solve its technical problems is: a potato planting disease and pest identification analysis method based on image processing, comprising: through unmanned aerial vehicle grid partition cruise shooting of potato planting field, panoramic images of each grid partition are obtained, and the panoramic images are compared with the reference images of the planting records of the same period, abnormal partitions are identified and their spatial positions are calibrated.

[0008] The spatial position of the abnormal partition is planned based on the unmanned aerial vehicle close-range cruise path, and the unmanned aerial vehicle is controlled to collect the current canopy image corresponding to the abnormal partition at close range.

[0009] The historical canopy image corresponding to the abnormal partition is called, the contribution score of each feature to the distinction of the disease and pest type is calculated according to the feature vector set of each disease and pest type at each time point, the optimal feature of the disease and pest type is distinguished based on the contribution score, and the feature difference matrix is constructed by comparing the feature difference values of each leaf in the current canopy image and the historical canopy image.

[0010] The standard image feature difference matrix of the potato disease and pest type is called, and the disease and pest type of the plant in the abnormal partition is judged by similarity matching.

[0011] If the disease and pest type of the plant in the abnormal partition cannot be judged, the tuber of the plant in the abnormal partition is sampled, the surface image and cross-section image of the tuber sample are collected, and the epidermal lesion morphological feature and cross-section decay area distribution feature are extracted.

[0012] The preset feature data set corresponding to each disease and pest type of the potato tuber is extracted, and the final disease and pest type of the abnormal partition is determined by comparison.

[0013] Compared with the prior art, the present application has the following beneficial effects: (1) the present application adopts unmanned aerial vehicle grid partition cruise shooting panoramic image, compares the reference image of the planting record of the same period pixel by pixel, screens the abnormal partition and calibrates the spatial position, realizes the global coverage detection of the potato planting field and the accurate positioning of the abnormal area, provides accurate spatial basis for the subsequent accurate identification and targeted prevention and control of diseases and pests, and improves the comprehensiveness of the detection of diseases and pests in large-scale planting fields.

[0014] (2) the present application calculates the contribution score of each feature to the distinction of the disease and pest type, selects the features with the average contribution score higher than the preset score threshold as the priority features, realizes the distinction difference of different disease and pest types, effectively eliminates redundant features, avoids artificial subjective errors, provides targeted feature support for the subsequent accurate identification of diseases and pests, and improves the reliability of the identification process.

[0015] (3) The application compares the feature difference values of each leaf in the current canopy image and the historical canopy image corresponding to the abnormal partition, and matches them with the standard image feature difference matrix to determine the plant disease and pest type of the abnormal partition, and through the comprehensive use of multi-dimensional and high-discriminative features, fine differentiation of different disease and pest types is realized, the recognition accuracy and stability are improved, and the problem of improper medication caused by misjudgment of disease type in the later stage is reduced.

[0016] (4) When the disease and pest type cannot be determined, the tuber sampling mechanism is triggered to collect the surface image and cross-section image of the tuber, extract the epidermal lesion shape feature and cross-section decay area distribution feature, determine the final disease and pest type of the abnormal partition, so as to realize comprehensive disease evaluation of the whole organ of the potato plant, ensure the comprehensiveness and integrity of the disease and pest identification, and reduce the yield loss risk caused by missed detection of tuber diseases. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0018] Figure 1 The figure is a schematic diagram of the method steps of the application.

[0019] Figure 2 The figure is a schematic diagram of the steps of identifying the abnormal partition and marking its spatial position in the application.

[0020] Figure 3 The figure is a schematic diagram of the steps of constructing the feature difference matrix in the application.

[0021] Figure 4 The figure is a schematic diagram of the preferred feature acquisition steps for distinguishing the disease and pest type in the application. DETAILED DESCRIPTION

[0022] Various exemplary embodiments of the application will now be described in detail with reference to the accompanying drawings. Note that the relative arrangement, numerical expressions and values of the components and steps set forth in these embodiments do not limit the scope of the application unless otherwise specifically stated. It should also be understood that the sizes of the various parts shown in the drawings are not drawn to scale for ease of description.

[0023] The following description of at least one example embodiment is merely exemplary in nature and is in no way intended to limit the application or its application or uses.

[0024] In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary, and not as a limitation. Other examples of the example embodiments can therefore have different values.

[0025] See Figure 1 As shown, the present application provides an image processing-based potato planting pest and disease identification analysis method, comprising: S1, the potato planting field is grid partitioned and patrolled by a drone, panoramic images of each grid partition are obtained, and the panoramic images are compared with reference images of the same period planting record to identify abnormal partitions and calibrate their spatial positions.

[0026] It should be noted that: in this embodiment, the application scenario is a large-scale potato planting field, and the grid partition is based on the crown area of a single potato plant, for example, the crown diameter of a conventional planting variety is about 0.6-0.7 meters, so the grid side length is set to 0.7 meters to form a square grid partition; the crown diameter of a dense planting variety is about 0.4-0.5 meters, so the grid side length is adjusted to 0.5 meters to ensure that a single grid accurately covers the crown of a single plant and avoids confusion of cross-plant features.

[0027] The same period planting record refers to the planting period record of the same potato planting field in the same period within recent years, planting the same potato variety and no occurrence of pests and diseases, to ensure the consistency of the growth environment and planting period of the reference image and the current image.

[0028] Preferably, in one embodiment of the present application, as Figure 2 As shown, the abnormal partition is identified and its spatial position is calibrated, specifically as follows: S11, the potato planting field is divided into several grid partitions according to the crown area of a single potato plant, panoramic images of each grid partition are obtained by drone cruising and shooting, and a mapping relationship between the panoramic images of each grid partition and the spatial position is established.

[0029] S12, the reference images of the same growth time as the current growth time are extracted from the potato planting management database in each same period planting record of the potato planting field, the panoramic images of each grid partition are compared with the reference images of each same period planting record pixel by pixel, and the color channel deviation value and the texture structure similarity are calculated.

[0030] It is required to be explained: the reference image is the image of the potato plant without disease and insect occurrence in the same growth time as the current growth time in each synchronous planting record. The color channel deviation value calculation adopts the absolute difference method, and the absolute gray difference values of all pixels of the panoramic image and the reference image are calculated in the pixel-by-pixel comparison. The mean value of the absolute gray difference values of the RGB channels is taken as the deviation value of the pixel, and the mean value of all pixel deviation values is taken as the color channel deviation value.

[0031] The texture structure similarity is calculated by using the gray level co-occurrence matrix algorithm, which is a prior art and will not be described in detail. The contrast, energy, entropy, correlation and other indicators in the texture structure feature are extracted, the texture feature vector is constructed, and the cosine similarity calculation formula is used to calculate the texture structure similarity of the panoramic image and the reference image. The texture structure similarity value ranges from 0 to 1, and the value closer to 1 indicates that the texture structure is more similar.

[0032] S13, screening the abnormal partition according to the color channel deviation value and the texture structure similarity, and extracting the center point position as the spatial position of the abnormal partition.

[0033] Preferably, the abnormal partition screening method is: first, the reference image of each synchronous planting record is subjected to gray scale processing, and the corresponding color channel feature value and texture structure feature value are extracted. The color channel feature value is the RGB channel gray value of all pixels in the reference image.

[0034] Secondly, the color channel feature value and the texture structure feature value of the reference image of each synchronous planting record are compared with those of other synchronous planting records, and all color channel feature difference values and texture structure feature similarities are obtained.

[0035] Then, all color channel feature difference values and texture structure feature similarities are subjected to outlier rejection, and the maximum color channel feature difference value and the minimum texture structure feature similarity are selected from the remaining color channel feature difference values and texture structure feature similarities, which are taken as the color channel reference deviation value and the texture structure reference similarity.

[0036] Finally, the grid partition that meets the color channel deviation value exceeding the color channel reference deviation value and the texture structure similarity being lower than the texture structure reference similarity at the same time is screened out, and is marked as an abnormal partition.

[0037] It should be noted that the outlier rejection adopts the 3σ principle, that is, the extreme data exceeding the mean value ± 3 times the standard deviation is rejected, the reference value is determined by rejecting the outliers through the 3σ principle, and the reliability of the reference value is ensured. There may be extreme characteristic values in the reference image due to temporary environmental interference during shooting. The 3σ principle effectively eliminates such interference data, avoids distortion of the reference value, and takes the maximum color channel deviation value and the minimum texture structure similarity of the remaining data as the reference value, defines the upper limit of the normal fluctuation, ensures that only the grid partition exceeding the normal fluctuation range will be judged as an abnormal partition, and effectively avoids the possibility of missing judgment.

[0038] The present application realizes the global coverage detection of potato planting field and the accurate positioning of abnormal area by adopting the unmanned aerial vehicle grid partition cruise shooting panoramic image, combining the benchmark image of the same period planting record, screening abnormal partition and calibrating spatial position, providing accurate spatial basis for subsequent accurate identification and targeted prevention and control of pests and diseases, and improving the comprehensiveness of large-scale planting field pest detection.

[0039] S2, based on the spatial position of the abnormal partition, planning the close-range cruise path of the unmanned aerial vehicle, controlling the unmanned aerial vehicle to collect the current canopy image corresponding to the abnormal partition.

[0040] The present embodiment plans the close-range cruise path of the unmanned aerial vehicle through the path optimization algorithm, which is the prior art, and the present embodiment will not be described in detail. At the same time, the unmanned aerial vehicle is controlled to collect the current canopy image corresponding to the abnormal partition, so as to ensure that the obtained canopy image can clearly reflect the detailed features of the leaves, and provide data support for disease and pest feature extraction and disease and pest type determination.

[0041] S3, calling the historical canopy image corresponding to the abnormal partition, calculating the contribution score of each feature to the distinction of disease and pest type according to the feature vector set of each disease and pest type at each time point, distinguishing the preferred feature of disease and pest type based on the contribution score, and comparing the feature difference value of each leaf in the current canopy image and the historical canopy image to construct a feature difference matrix.

[0042] Considering that the historical abnormal partition set does not contain the identified abnormal partition, the abnormal partition represents a new disease focus, and the feature change from normal to abnormal needs to be captured through the recent non-abnormal canopy image, therefore, the recent historical canopy image refers to the historical canopy image collected at the nearest time from the current time, and the canopy image which has not been judged as an abnormal partition.

[0043] Based on this, as shown in Figure 3 The content of constructing the feature difference matrix includes: first, calling the historical abnormal partition set from the potato planting management database, if the historical abnormal partition set does not contain the identified abnormal partition, calling the recent historical canopy image corresponding to the abnormal partition.

[0044] Secondly, according to the preferred features distinguishing the types of plant diseases and insect pests, all the preferred features in the current canopy image and the recent historical canopy images are extracted to construct feature vectors.

[0045] Finally, the Euclidean distances of the feature vectors of each leaf in the current canopy image and the recent historical canopy images are calculated, and the Euclidean distances are taken as feature difference values to form a feature difference matrix of each leaf.

[0046] It should be noted that the present embodiment is aimed at new lesions, and the feature vectors of the recent historical canopy images ensure that the abnormal partitions are compared with the features of the current canopy image. The feature changes in the initial stage of plant diseases and insect pests can be accurately captured, and the feature difference values of different leaves are calculated to avoid misplacement errors in the feature comparison process, solve the problem that the features of early plant diseases and insect pests are weak and difficult to identify, and provide a quantitative basis for rapid judgment of new lesions.

[0047] Considering that the set of historical abnormal partitions contains the identified abnormal partitions, which indicates that the abnormal partitions are recurrent or continuous lesions, and the feature development process of plant diseases and insect pests needs to be tracked, the historical time-series canopy image set needs to be called.

[0048] Therefore, the content of constructing the feature difference matrix further includes the following: first, if the set of historical abnormal partitions contains the identified abnormal partitions, the historical time-series canopy image set corresponding to the abnormal partitions is called, taking the first identified historical time point as the starting point and the current time point as the ending point.

[0049] Then, based on the feature vectors of the historical time-series canopy image set and the current canopy image, the feature difference values of each leaf in each adjacent time-series canopy image are calculated to form a feature difference matrix of each leaf.

[0050] It should be noted that the present embodiment is aimed at recurrent or continuous lesions, and the historical time-series canopy image set can completely capture the feature changes of plant diseases and insect pests from occurrence, development to spread. The feature comparison of adjacent time-series canopy images can reflect the development rate and severity of plant diseases and insect pests, and provide a basis for judging the severity of plant diseases.

[0051] Preferably, as shown in Figure 4 the preferred feature acquisition method for distinguishing the types of plant diseases and insect pests is as follows: first, the leaf images collected at each time point in each planting record corresponding to each type of plant diseases and insect pests are called from the potato planting management database, and each feature of the leaf images is extracted.

[0052] Secondly, the vector set of each feature corresponding to each time point of each type of plant diseases and insect pests is constructed, and the within-class mean vector and the within-class scatter matrix in the vector set of each feature are calculated.

[0053] In the third step, based on the within-class mean vector, Mahalanobis distances between different disease and pest types corresponding to each feature are calculated, and the Mahalanobis distances between different disease and pest types are subjected to ratio operation with the sum of the traces of the within-class scatter matrices corresponding to the types, to obtain the between-class separability criterion function values.

[0054] In the fourth step, the between-class separability criterion function values corresponding to each feature between different disease and pest types are weighted and summed, to obtain the contribution score of each feature to the disease and pest type distinction.

[0055] In the fifth step, the contribution score of each feature to the disease and pest type distinction at each time point is counted, the mean value of the contribution score corresponding to each feature is obtained, and the features with the mean value of the contribution score higher than a preset score threshold are selected as the priority features.

[0056] In an embodiment of the present application, the disease and pest types of the potato plants include, but are not limited to, early blight, late blight, black scab and bacterial wilt, and the features extracted from the leaf images include, but are not limited to, color moment mean, texture contrast, leaf circularity, lesion area ratio and leaf distribution density.

[0057] The weight setting between different disease and pest types is determined according to the normalized proportion of the planting record number of different disease and pest types, and in an embodiment, for example, when the planting record number corresponding to the early blight is 30, the planting record number corresponding to the late blight is 20, the planting record number corresponding to the black scab is 25 and the planting record number corresponding to the bacterial wilt is 25, the weight between the early blight and the late blight is .

[0058] The preset score threshold has a value range of 0.6-0.8, and the greater the number of disease and pest types in the planting area, the greater the value of the preset score threshold, and in an embodiment of the present application, for example, when the number of disease and pest types is less than 5, the preset score threshold is 0.6, and when the number of disease and pest types is greater than 10, the preset score threshold is 0.8.

[0059] According to the disease and pest complexity of the planting area, when the number of disease and pest types is ≤5, the preset score threshold is 0.6, when the number of disease and pest types is 5-10, the preset score threshold is 0.7, and when the number of disease and pest types is >10, the preset score threshold is 0.8.

[0060] It should be noted that the feature data needs to be standardized before the calculation of the Mahalanobis distance, and in order to eliminate the dimension influence, the trace of the within-class scatter matrix reflects the dispersion degree of the within-class samples, which is the sum of the diagonal elements of the within-class scatter matrix. The calculation of the within-class mean vector, the within-class scatter matrix and the Mahalanobis distance in the above content are all prior art, and the specific operation details are not described again.

[0061] The application scores the contribution of each feature to the type of plant diseases and insect pests, selects features with an average contribution score higher than a preset score threshold as priority features, realizes the differentiation of different types of plant diseases and insect pests, effectively eliminates redundant features, avoids manual subjective errors, provides targeted feature support for subsequent accurate identification of plant diseases and insect pests, and improves the reliability of the identification process.

[0062] S4, retrieve the standard image feature difference matrix of the type of potato plant diseases and insect pests, and judge the type of plant diseases and insect pests of the abnormal partition through similarity matching.

[0063] Considering that the occurrence and growth cycle of potato plant diseases and insect pests are strongly related, for example, late blight often occurs in the tuber formation period, and early blight is often seen in the present budding period, and the time sequence evolution characteristics of different plant diseases and insect pests are significantly different, the standard image feature difference matrix needs to be constructed according to different types of plant diseases and insect pests in different growth cycles.

[0064] Therefore, the standard image feature difference matrix of the type of potato plant diseases and insect pests is obtained in the following way: first, the standard leaf sample time sequence image set of each type of plant diseases and insect pests in each growth cycle is extracted from the potato plant disease and insect pest specimen library.

[0065] Secondly, according to the current growth time of the potato planting field, the standard leaf sample time sequence image set of each type of plant diseases and insect pests in the growth cycle of the potato planting field is determined.

[0066] Then, based on the interval length of adjacent canopy time sequence images in the feature difference matrix, each standard leaf sample time sequence image with the same interval length is selected from the standard leaf sample time sequence image set.

[0067] Finally, according to the feature vectors of each standard leaf sample time sequence image, the standard image feature difference matrix of each type of plant diseases and insect pests is formed.

[0068] It should be noted that the current growth time of the potato planting field is calculated by the sowing date of the planting management database.

[0069] In an example, for example, the interval length of adjacent canopy time sequence images is 24 hours, then from the standard leaf sample time sequence image set of each type of plant diseases and insect pests in the growth cycle of the potato planting field, the time sequence image with an interval length of 24 hours is selected.

[0070] Before judging the type of plant diseases and insect pests in this embodiment, a standardized standard image feature difference matrix needs to be constructed to ensure the consistency and adaptability of the standard image feature difference matrix and the feature difference matrix of the actual collected leaves, and provide data support for the accuracy of subsequent similarity matching.

[0071] It should be noted that the plant disease and pest type of the abnormal partition is determined by performing similarity calculation on the feature difference matrix of each leaf and the standard image feature difference matrix of each disease and pest type to obtain the feature difference matrix similarity of each leaf and each disease and pest type.

[0072] When the feature difference matrix similarity of any leaf and a certain disease and pest type is higher than the set similarity threshold and is the highest among the feature difference matrix similarities of all disease and pest types, the disease and pest type is determined as the plant disease and pest type of the abnormal partition.

[0073] When the feature difference matrix similarity of all leaves and each disease and pest type is lower than the set similarity threshold, it is determined that the plant disease and pest type of the abnormal partition cannot be determined.

[0074] In an embodiment of the present application, the similarity calculation adopts a cosine similarity algorithm, which can effectively quantify the feature matching degree of two matrices, and the similarity value ranges from 0 to 1. The closer the value is to 1, the higher the matching degree. This avoids subjective judgment errors and ensures the reliability of the determination result. The set similarity threshold is 0.8, which can be adjusted by the implementer according to the complexity of the planting area, and the adjustment range is 0.75-0.85.

[0075] The present application determines the plant disease and pest type of the abnormal partition by comparing the feature difference values of each leaf in the current canopy image and the historical canopy image corresponding to the abnormal partition with the standard image feature difference matrix, and performing similarity matching. Through the comprehensive use of multi-dimensional and high-discriminative features, fine differentiation of different disease and pest types is realized, the recognition accuracy and stability are improved, and the problem of improper medication caused by misjudgment of disease type in the later stage is reduced.

[0076] S5, if the plant disease and pest type of the abnormal partition cannot be determined, tuber sampling is performed on the plant in the abnormal partition, surface images and cross-section images of the tuber samples are collected, and epidermal lesion morphology features and cross-sectional decay area distribution features are extracted.

[0077] Considering that the tuber lesion morphology and decay distribution caused by different diseases and pests are significantly different, for example, the tuber lesions of late blight are mostly circular, and the decay spreads from the epidermis to the center; the tuber lesions of black scurf disease are irregular protrusions, and the decay is limited to the epidermis. Therefore, the feature extraction needs to consider both the morphology and the distribution to ensure that the features of different diseases and pests can be distinguished.

[0078] Therefore, the extraction method of the epidermal lesion morphology features and the cross-sectional decay area distribution features is as follows: S51, a non-destructive sampling tool is used to sample the tubers at the root of the potato plant corresponding to the abnormal partition, and the surface of the tuber sample is cleaned.

[0079] S52, adopt image acquisition equipment to shoot the surface image of the tuber sample after cleaning treatment, segment the surface image, extract the outline curve of the diseased area, and calculate the circularity, aspect ratio and concave-convex degree of the diseased area, which are taken as the epidermal lesion morphology characteristics.

[0080] In an embodiment of the present application, the threshold segmentation algorithm based on the HSV color space is used for the segmentation of the diseased area, which is a prior art, and thus will not be described in detail.

[0081] The formula for calculating the circularity is: Wherein S is the area of the diseased area, L is the perimeter of the diseased outline, Pi is the ratio of the circumference of a circle to its diameter, The circularity is a value ranging from 0 to 1, and the closer to 1 indicates that the lesion is closer to a circle.

[0082] The aspect ratio is the ratio of the longest axis length to the shortest axis length of the diseased outline.

[0083] The calculation method of the concave-convex degree is as follows: the difference between the convex hull perimeter of the diseased outline and the diseased outline perimeter is obtained, and the ratio of the difference to the convex hull perimeter of the diseased outline is taken as the concave-convex degree, wherein the greater the concave-convex degree indicates that the diseased outline is more irregular.

[0084] S53, vertically section the tuber sample to obtain a cross-sectional image, perform color space conversion on the cross-sectional image, separate the rotten area and the healthy area, extract the rotten area area ratio and the distance between the rotten area and the tuber center, and take them as the cross-sectional rotten area distribution characteristics.

[0085] In an embodiment of the present application, after the RGB to Lab color space conversion, the K-means clustering algorithm based on the a channel and the b channel is used, the number of clusters is set to 2, corresponding to the healthy area and the rotten area respectively, and the cluster centers are preset through the standard sample library, which is a prior art and thus will not be described in detail.

[0086] S6, extract the preset feature data set of each disease and pest type corresponding to the potato tuber, and compare and determine the final disease and pest type of the abnormal partition.

[0087] It should be noted that the preset feature data set of each disease and pest type corresponding to the potato tuber is stored in the tuber feature sub-library of the potato disease and pest specimen library, the feature data set of each disease and pest type is constructed through a plurality of standard tuber samples, and the preset feature range of each disease and pest type is the mean value ± 2 times the standard deviation of each feature data, which conforms to the normal distribution statistical law and covers more than 95% of the standard sample features, thereby avoiding missed judgment caused by too narrow range and preventing the decline of the discrimination caused by too wide range, laying a solid foundation for subsequent accurate matching.

[0088] Specifically, the final disease type of the abnormal partition is determined as follows: first, the preset range of the epidermal lesion morphology feature and the preset range of the cross-sectional decay area distribution feature are extracted from the preset feature data set corresponding to each disease type of the potato tuber.

[0089] Then, the number of features within the preset range of the epidermal lesion morphology feature and the preset range of the cross-sectional decay area distribution feature is matched by combining the epidermal lesion morphology feature and the cross-sectional decay area distribution feature of the tuber sample.

[0090] Finally, the disease type with the largest number of features is selected as the final disease type of the abnormal partition.

[0091] When the disease type cannot be determined, the tuber sampling mechanism is triggered to collect the surface image and cross-sectional image of the tuber, the epidermal lesion morphology feature and the cross-sectional decay area distribution feature are extracted, and the final disease type of the abnormal partition is determined, so as to realize comprehensive disease evaluation of the whole organ of the potato plant, ensure the comprehensiveness and integrity of disease identification, and reduce the yield loss risk caused by missed detection of tuber diseases.

[0092] The above embodiments can be realized by software, hardware, firmware or any combination thereof, in whole or in part. When realized by software, the above embodiments can be realized in the form of a computer program product in whole or in part.

[0093] Those skilled in the art can realize that the modules and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized by hardware or software depends on the specific application and design constraints of the technical solutions. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0094] In addition, the functional modules in each embodiment of the present application can be integrated in one processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0095] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any skilled person in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0096] Finally, the above only is the preferred embodiment of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for identifying and analyzing potato planting diseases and pests based on image processing, characterized in that, include: By using drones to conduct grid-based patrols and photography of potato fields, panoramic images of each grid area are obtained and compared with baseline images of planting records from the same period to identify abnormal areas and mark their spatial locations. Based on the spatial location of abnormal partitions, plan the close-range cruise path of the UAV and control the UAV to collect the current canopy image corresponding to the abnormal partition at close range; Retrieve historical canopy images corresponding to abnormal partitions, calculate and evaluate the contribution score of each feature to the differentiation of pest and disease types based on the feature vector set of each pest and disease type at each time point, select the preferred features to differentiate pest and disease types based on the contribution score, compare the feature difference values ​​of each leaf in the current canopy image and the historical canopy image, and construct a feature difference matrix. The specific construction content includes: retrieving the historical abnormal partition set from the potato planting management database; if the historical abnormal partition set does not contain the identified abnormal partition, then retrieving the recent historical canopy image corresponding to the abnormal partition. Based on the preferred features for distinguishing pest and disease types, extract all preferred features from the current canopy image and recent historical canopy images, and construct a feature vector; Calculate the Euclidean distance between the feature vectors of each leaf in the current canopy image and the recent historical canopy images, and use the Euclidean distance as the feature difference value to form the feature difference matrix of each leaf; If the historical abnormal partition set contains the identified abnormal partition, then the historical time point of the first identification of the abnormal partition is taken as the starting point and the current time point is taken as the ending point, and the historical time series canopy image set corresponding to the abnormal partition is retrieved. Based on the feature vectors of historical canopy images and current canopy images, the feature difference values ​​of each leaf in each adjacent canopy image are calculated to form the feature difference matrix of each leaf. Retrieve the standard image feature difference matrix of potato disease and pest types, and determine the type of plant disease and pest in abnormal partitions by similarity matching; If the type of plant disease or pest in the abnormal zone cannot be determined, tuber samples are taken from the plants in the abnormal zone. Surface and cross-sectional images of the tuber samples are collected to extract the morphological characteristics of epidermal lesions and the distribution characteristics of rotten areas in the cross-section. Extract the preset feature datasets corresponding to each type of disease and pest in potato tubers, and compare them to determine the final type of disease and pest in abnormal partitions.

2. The method for identifying and analyzing potato planting diseases and pests based on image processing according to claim 1, characterized in that: The process of identifying abnormal partitions and marking their spatial locations is as follows: The potato field was divided into several grid zones according to the canopy area of ​​a single potato plant. Panoramic images of each grid zone were obtained by drone patrol and photography, and a mapping relationship between the panoramic images of each grid zone and its spatial location was established. The baseline images with the same growth time in each planting record of potato fields at the same time are extracted from the potato planting management database. The panoramic images of each grid partition are compared pixel by pixel with the baseline images of each planting record at the same time, and the color channel deviation value and texture structure similarity are calculated. Abnormal partitions are filtered based on color channel deviation values ​​and texture structure similarity, and their center point positions are extracted as the spatial locations of the abnormal partitions.

3. The method for identifying and analyzing potato planting diseases and pests based on image processing according to claim 2, characterized in that: The abnormal partition filtering method is as follows: The baseline images of each planting record from the same period were processed in grayscale, and the corresponding color channel feature values ​​and texture structure feature values ​​were extracted. By comparing the color channel feature values ​​and texture structure feature values ​​of each planting record with the reference image of other planting records from the same period, the difference of all color channel features and the similarity of texture structure features are obtained. Outliers are removed from all color channel feature differences and texture structure feature similarities. The largest color channel feature difference and the smallest texture structure feature similarity are selected from the remaining color channel feature differences and texture structure feature similarities and used as the color channel reference deviation value and texture structure reference similarity value. Mesh partitions that simultaneously meet the criteria of color channel deviation exceeding the color channel reference deviation and texture structure similarity being lower than the texture structure reference similarity are identified and marked as abnormal partitions.

4. The method for identifying and analyzing potato planting diseases and pests based on image processing according to claim 1, characterized in that: The preferred feature for distinguishing pest and disease types is obtained as follows: Leaf images collected at various time points from each planting record corresponding to each type of pest and disease were retrieved from the potato planting management database, and various features of the leaf images were extracted. Construct vector sets for each feature corresponding to each type of pest and disease at each time point, and calculate the intra-class mean vector and intra-class scatter matrix in the vector set of each feature; Based on the intra-class mean vector, the Mahalanobis distance between different pest and disease types corresponding to each feature is calculated. The Mahalanobis distance between different pest and disease types is then compared with the sum of the traces of the intra-class scatter matrices of the corresponding types to obtain the inter-class separability criterion function value. The inter-class separability criterion function values ​​corresponding to different pest and disease types for each feature are weighted and summed to obtain the contribution score of each feature to the differentiation of pest and disease types. The contribution scores of each feature to the differentiation of pest and disease types are statistically analyzed at each time point. The average contribution score of each feature is obtained, and features with an average contribution score higher than a preset score threshold are selected as priority features.

5. The method for identifying and analyzing potato planting diseases and pests based on image processing according to claim 1, characterized in that: The standard image feature difference matrix for the potato disease and pest types is obtained as follows: Extract time-series image sets of standard leaf samples for each type of disease and pest during each growth cycle from the potato disease and pest specimen bank; Based on the current growth time of the potato field, determine the time series image set of standard leaf samples for each type of pest and disease within the growth cycle of the potato field. Based on the interval duration of adjacent canopy time series images in the feature difference matrix, time series images of standard leaf samples with the same interval duration are selected from the set of standard leaf sample time series images; Based on the feature vectors of the time-series images of each standard leaf sample, a standard image feature difference matrix is ​​formed for each type of pest and disease.

6. The method for identifying and analyzing potato planting diseases and pests based on image processing according to claim 5, characterized in that: The determination of plant disease and pest types in abnormal zones specifically includes: The similarity calculation is performed between the feature difference matrix of each leaf and the feature difference matrix of the standard image of each disease and pest type to obtain the similarity between the feature difference matrix of each leaf and each disease and pest type. When the similarity of the feature difference matrix between any leaf and a certain type of pest or disease is higher than the set similarity threshold and is the highest among all types of pests and diseases, then that type of pest or disease is regarded as the plant pest or disease type in the abnormal partition. If the similarity of the feature difference matrix of all leaves with each type of disease and pest is lower than the set similarity threshold, then the plant disease and pest type of the abnormal partition cannot be determined.

7. The method for identifying and analyzing potato planting diseases and pests based on image processing according to claim 1, characterized in that: The extraction methods for the morphological characteristics of the epidermal lesions and the distribution characteristics of the necrotic areas in the cross-section are as follows: Tubers were sampled from the roots of potato plants in the abnormal zone using non-destructive sampling tools, and the tuber samples were then surface-cleaned. The surface images of tuber samples after cleaning were captured using an image acquisition device. The lesion areas of the surface images were segmented, the contour curves of the lesion areas were extracted, and the roundness, aspect ratio, and concavity of the lesion areas were statistically analyzed and used as the morphological characteristics of epidermal lesions. The tuber samples were vertically sectioned to obtain cross-sectional images. The cross-sectional images were then color space converted to separate rotten and healthy areas. The area ratio of rotten areas and the distance between rotten areas and the center of the tuber were extracted and used as the distribution characteristics of rotten areas in the cross-section.

8. The method for identifying and analyzing potato planting diseases and pests based on image processing according to claim 7, characterized in that: The final pest and disease type for determining the abnormal partition is: Extract the preset range of epidermal lesion morphology features and the preset range of cross-sectional rot area distribution features from the preset feature dataset corresponding to each type of disease and pest in potato tubers; By combining the morphological characteristics of epidermal lesions and the distribution characteristics of rotten areas in the cross-section of tuber samples, the number of features that fall within the preset range of epidermal lesion morphological characteristics and the preset range of rotten area distribution characteristics in the cross-section is obtained. The pest and disease type with the most features is selected as the final pest and disease type for the abnormal partition.

Images