Method for the cultivation management of crops

Abstract

The application discloses a kind of crop planting management methods.It first obtains the growth state image of monitored crop by camera acquisition, then, image feature extraction is carried out to the growth state image to obtain growth state shallow feature vector and growth state deep feature vector, then, based on the growth state shallow feature vector and the growth state deep feature vector, determine the recommended pesticide type.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of intelligent management, and more particularly, to a planting management method of crops. BACKGROUND

[0002] Pest and disease of crops is one of the common problems in agricultural production, and timely discovery and use of targeted pesticides can effectively solve the problem of pest and disease of crops, thereby improving agricultural income.

[0003] Specifically, in the growth process of crops, different pest and disease threats will be encountered, and different pest and disease corresponds to different pesticide control methods. For example, crops in the seedling stage are weak in resistance to pest and disease and are easily attacked by pest and disease, at which time pesticides with broad-spectrum insecticidal and fungicidal functions are needed to prevent and control common pest and disease, and in the growth period, specific diseases may occur due to certain specific environment, and targeted pesticides need to be selected for treatment.

[0004] Usually, before pesticide treatment, manual inspection or fixed-point sampling is needed, which is time-consuming and labor-intensive and difficult to cover large areas of farmland. Therefore, an optimized planting management scheme of crops is expected. SUMMARY

[0005] Therefore, the present disclosure provides a planting management method of crops, which can use a camera to collect growth state images of monitored crops and extract growth state feature information about the monitored crops from the images, so as to intelligently recommend pesticides suitable for the current growth stage, thereby improving the resistance of crops and agricultural income.

[0006] According to an aspect of the present disclosure, a planting management method of crops is provided, which includes:

[0007] obtaining growth state images of monitored crops collected by a camera;

[0008] performing image feature extraction on the growth state images to obtain a growth state shallow feature vector and a growth state deep feature vector; and

[0009] determining a recommended pesticide type based on the growth state shallow feature vector and the growth state deep feature vector.

[0010] According to the embodiment of the present disclosure, firstly, a growth state image of a monitored crop is acquired by a camera, then, image feature extraction is performed on the growth state image to obtain a growth state shallow feature vector and a growth state deep feature vector, and then, a recommended pesticide type is determined based on the growth state shallow feature vector and the growth state deep feature vector. In this way, the growth state image of the monitored crop can be acquired by the camera, and the growth state feature information about the monitored crop is extracted therefrom, so as to intelligently recommend the pesticide suitable for the current growth stage, thereby improving the resistance of the crop and the agricultural income.

[0011] Other features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the drawings. BRIEF DESCRIPTION OF DRAWINGS

[0012] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the present disclosure and serve to explain the principles of the present disclosure.

[0013] Figure 1 A flowchart of a crop planting management method according to an embodiment of the present disclosure is shown.

[0014] Figure 2 An architectural schematic diagram of a crop planting management method according to an embodiment of the present disclosure is shown.

[0015] Figure 3 A flowchart of sub-step S120 of a crop planting management method according to an embodiment of the present disclosure is shown.

[0016] Figure 4 A flowchart of sub-step S130 of a crop planting management method according to an embodiment of the present disclosure is shown.

[0017] Figure 5 A flowchart of sub-step S132 of a crop planting management method according to an embodiment of the present disclosure is shown.

[0018] Figure 6 A block diagram of a crop planting management system according to an embodiment of the present disclosure is shown.

[0019] Figure 7 An application scenario diagram of a crop planting management method according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0020] With reference to the drawings, the technical solutions in the embodiments of the present disclosure will be clearly and completely described, obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all the embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by a person of ordinary skill in the art without creative labor also belong to the protection scope of the present disclosure.

[0021] As shown in the present disclosure and claims, unless the context clearly indicates otherwise, "one", "a", "an", and / or "the" do not specify a single number, but can also include a plurality. Generally, the terms "comprise" and "include" only indicate the inclusion of the steps and elements explicitly identified, and these steps and elements do not constitute an exclusive list, and the method or device can also include other steps or elements.

[0022] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference signs in the drawings represent functionally identical or similar elements. Although various aspects of the embodiments are shown in the drawings, the drawings are not necessarily drawn to scale unless specifically indicated.

[0023] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the specific embodiments below. A person of ordinary skill in the art should understand that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, elements and circuits that are well known to those skilled in the art are not described in detail in order to highlight the main ideas of the present disclosure.

[0024] To solve the above technical problems, the technical concept of the present disclosure is to use a camera to collect growth state images of monitored crops, and to extract growth state feature information about the monitored crops from the images, so as to intelligently recommend pesticides suitable for the current growth stage, thereby improving the resistance of crops and agricultural income.

[0025] Figure 1 A flowchart of a crop planting management method according to an embodiment of the present disclosure is shown. Figure 2 An architectural schematic diagram of a crop planting management method according to an embodiment of the present disclosure is shown. As Figure 1 and Figure 2 As shown in the crop planting management method according to the embodiment of the present disclosure, the method comprises the steps of: S110, acquiring growth state images of monitored crops collected by a camera; S120, performing image feature extraction on the growth state images to obtain a growth state shallow feature vector and a growth state deep feature vector; and S130, determining a recommended pesticide type based on the growth state shallow feature vector and the growth state deep feature vector.

[0026] In the step S110, a common digital camera or a professional agricultural camera can be selected to capture the growth state image of the monitored crop. Specifically, a camera with high resolution and good image quality should be selected to ensure that a clear image can be obtained. Agricultural cameras usually have better adaptability and stability, and can provide more accurate images under different environmental conditions. The camera should be installed in a suitable position to obtain the best view. The camera should be able to cover the entire crop area and capture the growth state changes of the plants. At the same time, the environment where the camera is located should have appropriate lighting conditions to obtain clear images. Too strong or too weak light can cause image quality to decrease and affect the accuracy of feature extraction. Further, according to specific requirements, the shooting frequency and time interval of the camera can be set. Continuous shooting or timed shooting can be selected to capture the growth state changes of the crop. The images can be stored on a local device or a remote server for subsequent processing and analysis. In other words, selecting an appropriate camera and setting it according to actual needs can effectively capture the growth state image of the monitored crop, which will provide a reliable data basis for subsequent image feature extraction and pesticide recommendation.

[0027] More specifically, in step S120, as shown in Figure 3 the growth state image is subjected to image feature extraction to obtain a growth state shallow feature vector and a growth state deep feature vector, including: S121, the growth state image is subjected to image preprocessing to obtain a denoised growth state image; S122, a growth state shallow feature map is extracted from the denoised growth state image; S123, a growth state deep feature map is extracted from the growth state shallow feature map; and S124, the growth state shallow feature map and the growth state deep feature map are respectively subjected to dimension reduction processing to obtain the growth state shallow feature vector and the growth state deep feature vector.

[0028] Accordingly, in the technical solution of the present disclosure, the growth state image of the monitored crop captured by the camera is first obtained, and the growth state image is subjected to image preprocessing to obtain a denoised growth state image. In the embodiments of the present disclosure, the growth state image is subjected to image denoising to obtain a denoised growth state image. Since the environment where the camera captures the image is located in a farmland, it is affected by harsh weather such as wind, rain, and dust flying, and the obtained growth state image usually has noise or a large amount of interference information. In addition, the details of the leaves of the crops may be blocked or overlapped, causing confusion and blur in the image. Through denoising, such interference can be effectively reduced, providing an important data source for subsequent image feature extraction.

[0029] Correspondingly, in one specific example, the growth state image is subjected to image preprocessing to obtain a denoised growth state image, which includes: subjecting the growth state image to image denoising processing to obtain the denoised growth state image. It should be understood that image denoising processing is a technical method for improving image quality and clarity by reducing noise and interference information in the image. In crop growth state image processing, image denoising processing can play the following roles: removing noise and interference, there are wind and rain, dust flying, and other adverse weather conditions in the farmland environment, which can cause noise and interference in the image. Through denoising processing, these noise and interference can be removed, improving the clarity and readability of the image; improving image quality, denoising processing can reduce noise and artifacts in the image, making the image clearer and the details more prominent, thereby improving the quality of the image; improving the effect of subsequent processing, denoising processing can provide more accurate data sources for subsequent image feature extraction and analysis, and the image after removing noise and interference is easier to be processed by algorithms and models, which can improve the accuracy and effect of subsequent processing. Specific image denoising processing can be performed in the following ways: 1. Mean filtering, which smoothes the image by calculating the average value of the surrounding neighborhood pixels, reducing the influence of noise; 2. Median filtering, which smoothes the image by calculating the median value of the surrounding neighborhood pixels, and has good removal effect on certain types of noise such as salt and pepper noise; 3. Gaussian filtering, which uses a Gaussian function to smooth the image, which can effectively reduce high-frequency noise; 4. Wavelet denoising, which uses wavelet transform to decompose the image into multiple frequency bands, and then reconstructs the image after threshold processing of the coefficients of different frequency bands, which can effectively remove noise. These denoising methods can be used alone or in combination, and the appropriate method is selected for image denoising processing according to the specific situation.

[0030] Considering that the convolutional neural network model has a natural advantage in image feature extraction, but as the number of layers of the convolutional neural network model increases, the shallow feature information of the image, such as shape, color and texture, will be eroded, and these shallow feature information can be used to represent the growth state of crops. Therefore, in the technical solution of the present disclosure, it is expected to retain the shallow feature information of the image and fuse the shallow feature information with the deep feature information to enrich the expression of the features.

[0031] Specifically, in the technical solution of the present disclosure, the growth state image after noise reduction is first passed through a shallow feature extractor based on a first convolutional neural network model to obtain a growth state shallow feature map; then, the growth state shallow feature map is passed through a deep feature extractor based on a second convolutional neural network model to obtain a growth state deep feature map; further, the growth state shallow feature map and the growth state deep feature map are respectively unfolded into a growth state shallow feature vector and a growth state deep feature vector; and then, a cascade function is used to fuse the growth state shallow feature vector and the growth state deep feature vector to obtain a classification feature vector.

[0032] Accordingly, in one specific example, the growth state shallow feature map is extracted from the denoised growth state image, including: passing the denoised growth state image through a shallow feature extractor based on a first convolutional neural network model to obtain the growth state shallow feature map; the growth state deep feature map is extracted from the growth state shallow feature map, including: passing the growth state shallow feature map through a deep feature extractor based on a second convolutional neural network model to obtain the growth state deep feature map. It should be understood that the convolutional neural network (Convolutional Neural Network, CNN) is a deep learning model, which is particularly suitable for processing and analyzing two-dimensional data such as images and videos. It is constructed through multiple convolutional layers and pooling layers, which can automatically learn the features in the image, and achieve good results in tasks such as classification, target detection, image generation, etc. The main features of the convolutional neural network model include: 1. Local perception, the convolutional layer uses a convolution kernel to convolve the input image, and extracts local features through a local receptive field, thereby capturing the spatial structure information of the image; 2. Parameter sharing, in the convolutional layer, the parameters of the convolution kernel are shared throughout the image, which can greatly reduce the parameter amount of the model and improve the training efficiency of the model; 3. Pooling operation, the pooling layer is used for downsampling, which reduces the size of the feature map through pooling operations (such as max pooling, average pooling) on local regions, and extracts more robust features; 4. Multi-layer stacking, the convolutional neural network is usually composed of multiple convolutional layers, pooling layers and fully connected layers, through multi-layer stacking, more abstract and high-level features can be gradually extracted. Further, the shallow feature extractor based on the first convolutional neural network model and the deep feature extractor based on the second convolutional neural network model differ in the following aspects: 1. Number of layers and complexity, the second convolutional neural network model has a deeper number of layers and a more complex structure than the first convolutional neural network model, the deep feature extractor contains more convolutional layers, pooling layers and fully connected layers, and can better capture the high-level semantic information of the image; 2. Feature abstraction ability, the deep feature extractor has a stronger feature abstraction ability than the shallow feature extractor, through the stacking of multiple convolutional layers and nonlinear activation functions, the deep feature extractor can gradually extract more abstract and semantic features, thereby better representing the target objects or structures in the image; 3. Parameter amount and computational complexity, since the deep feature extractor has more layers and parameters, its parameter amount and computational complexity are usually higher, which means that training and reasoning the deep feature extractor requires more computational resources and time; 4. Transfer learning ability, the deep feature extractor usually has better transfer learning ability, since the deep feature extractor can learn more general feature representations, it can be applied to other related image processing tasks through fine-tuning or feature extraction on different tasks and datasets.In general, deep feature extractors have stronger expressive and feature abstraction capabilities than shallow feature extractors, but also require more computational resources and time for training and inference.

[0033] Correspondingly, in one specific example, the dimensionality reduction processing of the growth state shallow feature map and the growth state deep feature map to obtain the growth state shallow feature vector and the growth state deep feature vector respectively includes: unfolding the growth state shallow feature map and the growth state deep feature map into the growth state shallow feature vector and the growth state deep feature vector respectively. It is worth mentioning that the unfolding of the growth state shallow feature map and the growth state deep feature map into the feature vector is achieved by arranging the pixel values of the image into a one-dimensional vector in a certain order. In one example of the present disclosure, the specific steps are as follows: 1. Growth state shallow feature map unfolding: in the order of from left to right and from top to bottom, the value of each pixel is added to the feature vector in turn until the entire feature map is unfolded into a one-dimensional vector; 2. Growth state deep feature map unfolding: similarly, in the order of from left to right and from top to bottom, the value of each pixel is added to the feature vector in turn until the entire feature map is unfolded into a one-dimensional vector. The purpose of unfolding the growth state shallow feature map and the growth state deep feature map into the feature vector is to convert the image data into a numerical vector that can be used by machine learning algorithms. By converting the image into a feature vector, the image processing problem can be converted into a common numerical calculation problem such as classification, regression, etc.; by calculating the distance or similarity between feature vectors, image searching and matching can be achieved, such as finding similar images to a given image in a large-scale image database; after converting the image into a feature vector, dimensionality reduction techniques can be used to map the high-dimensional feature vector to a low-dimensional space, thereby realizing image visualization and analysis. In other words, unfolding the growth state shallow feature map and the growth state deep feature map into the feature vector can facilitate the application of image data to machine learning algorithms and realize image searching, visualization and analysis, etc.

[0034] Correspondingly, in one specific example, as Figure 4 shown, based on the growth state shallow feature vector and the growth state deep feature vector, determining the recommended pesticide type includes: S131, using a cascade function to fuse the growth state shallow feature vector and the growth state deep feature vector to obtain a classification feature vector; S132, performing information gain on the classification feature vector to obtain an optimized classification feature vector; and S133, passing the optimized classification feature vector through a classifier to obtain a classification result, which is used to represent a recommended pesticide type label.

[0035] More specifically, the growth state shallow feature vector and the growth state deep feature vector are fused using a concatenation function to obtain a classification feature vector, including: the growth state shallow feature vector and the growth state deep feature vector are fused using a concatenation function as follows to obtain the classification feature vector; wherein the concatenation function is:

[0036] f(X i ,X j )=Relu(W f [θ(X i ),φ(X j )])

[0037] Wherein W f , θ(X i ) and φ(X j ) all represent point convolution on input, Relu is an activation function, [] represents a splicing operation, X i is a feature value of each position in the growth state shallow feature vector, and X j is a feature value of each position in the growth state deep feature vector.

[0038] Then, the optimized classification feature vector is passed through a classifier to obtain a classification result, which is used to represent a recommended pesticide type label. It is worth mentioning that the pesticide type label is divided according to different classification topics, which can include the following: disease type: according to the specific disease type suffered by the crop, the pesticide type label can be the corresponding disease name, such as "powdery mildew", "black spot", "aphid", etc.; pesticide function: according to the different action modes and effects of pesticides, the pesticide type label can be the corresponding function description, such as "fungicide", "insecticide", "herbicide", etc.; pesticide composition: according to the main component composition of the pesticide, the pesticide type label can be the corresponding chemical component name, such as "cypermethrin", "imidacloprid", "phenol formate", etc.; use mode: according to the use mode and application method of the pesticide, the pesticide type label can be the corresponding use mode description, such as "spray type", "irrigation type", "smoke type", etc. The selection of these pesticide type labels depends on the specific application scene and demand, and can be selected and defined according to the actual situation.

[0039] It should be understood that the role of the classifier is to learn classification rules and classifiers using given classes and known training data, and then classify (or predict) unknown data. Logistics, SVM, etc. are commonly used to solve binary classification problems. For multi-class classification, logistics or SVM can also be used, but multiple binary classifications are needed to form multi-classification, which is prone to errors and low efficiency. Common multi-classification methods include Softmax classification function.

[0040] A classifier is a machine learning model that classifies input data into different categories or labels. It can learn classification rules and patterns from known training data and apply these rules to unknown data for classification prediction. The application of classifiers in crop growth state image processing can help identify the types of diseases suffered by crops and recommend appropriate pesticide types. By classifying the optimized classification feature vector, the classifier can learn the differences and features between different categories based on existing training data, and then apply these learned rules to new image data to predict its category or label. Common classifiers include logistic regression, support vector machine (SVM), decision tree, random forest, and naive Bayes. For binary classification problems, logistic regression and SVM are commonly used methods. For multi-classification problems, logistic regression or SVM can be used in combination with one-vs-rest or one-vs-one strategies to combine multiple binary classification tasks. In addition, the Softmax classification function is also a common multi-classification method, which can map the input feature vector to a probability distribution of different categories. The role of the classifier is to classify and predict unknown data, helping to understand and interpret data, and making decisions and recommendations based on classification results. In crop growth state image processing, classifiers can identify diseases and recommend pesticides based on crop characteristics and image information, which can help manage and make decisions in agricultural production.

[0041] Correspondingly, in one possible implementation, the optimized classification feature vector is input into a classifier to obtain a classification result, the classification result being used to represent a recommended pesticide type label, including: performing full connection coding on the optimized classification feature vector using a full connection layer of the classifier to obtain an encoded classification feature vector; and inputting the encoded classification feature vector into a Softmax classification function of the classifier to obtain the classification result.

[0042] As you can understand, the Softmax classification function is a commonly used multi-class classification function that transforms an input vector into a vector representing the probabilities of each class. The Softmax function can convert any real-valued vector into a probability distribution such that the sum of the probabilities of each class is 1. When using the Softmax classification function, the encoded classification feature vector is taken as input, encoded by a fully connected layer to obtain a new vector, which is then input into the Softmax classification function. Through the calculation of the Softmax function, a vector representing the probability of each class is obtained. Finally, the class corresponding to the maximum value in the probability vector can be used as the classification result, representing the recommended pesticide type label.

[0043] It is worth mentioning that, in a specific example, such as Figure 5 As shown, the process of applying information gain to the classification feature vector to obtain an optimized classification feature vector includes: S1321, performing forward propagation information-preserving fusion on the shallow growth state feature vector and the deep growth state feature vector to obtain a corrected feature vector; S1322, performing linear interpolation on the corrected feature vector to obtain an adjusted corrected feature vector, wherein the adjusted corrected feature vector has the same length as the classification feature vector; and S1323, performing a dot product weighted sum of the adjusted corrected feature vector and the classification feature vector to obtain the optimized classification feature vector.

[0044] It should be understood that linear interpolation is used to adjust the values of the correction feature vector, and dot product weighting is used to fuse the adjusted correction feature vector with the classification feature vector to obtain the optimized classification feature vector. Specifically, linear interpolation is an interpolation method used to estimate the value of an unknown data point between known data points. In this specific example, linear interpolation is used to adjust the values of the correction feature vector. Through linear interpolation, the feature values at unknown positions can be estimated based on the known values of the correction feature vector to obtain a smoother and continuous feature vector. Linear interpolation can be achieved by calculating the linear relationship between two known data points. Dot product weighting is a vector operation used to weight and fuse two vectors. In this specific example, dot product weighting is used to fuse the adjusted correction feature vector with the classification feature vector to obtain the optimized classification feature vector. The operation of dot product weighting is to multiply the corresponding elements of two vectors and then add the products to obtain a scalar value. Through dot product weighting, each element of the adjusted correction feature vector can be weighted and fused with the corresponding element of the classification feature vector to obtain a comprehensive feature vector. Through the operations of linear interpolation and dot product weighting, the correction feature vector can be adjusted and fused to obtain the optimized classification feature vector. Such optimization process can improve the expression ability and classification performance of the feature vector, thereby better representing the growth state of crops and providing more accurate feature input for subsequent classification tasks.

[0045] Further, in the technical solution of the present disclosure, the growth state shallow feature vector and the growth state deep feature vector respectively express the shallow and deep image semantic features of the growth state image, and therefore, due to the difference in feature expression depth, the image semantic feature distributions of the growth state shallow feature vector and the growth state deep feature vector are not completely aligned.

[0046] In this way, when the growth state shallow feature vector and the growth state deep feature vector are fused to obtain the classification feature vector using the cascade function, the respective misaligned image semantic feature distributions of the growth state shallow feature vector and the growth state deep feature vector will cause information loss during the forward propagation of the model when performing dot convolution and activation operations via the cascade function, affecting the accuracy of the classification result obtained by the classifier through the classification feature vector. Based on this, the applicant of the present disclosure performs forward propagation information preservation fusion on the growth state shallow feature vector, denoted as V1, and the growth state deep feature vector, denoted as V2, to obtain a correction feature vector V'.

[0047] Correspondingly, in one specific example, the forward propagation information retention fusion of the growth state shallow layer feature vector and the growth state deep layer feature vector to obtain a corrected feature vector comprises: performing forward propagation information retention fusion of the growth state shallow layer feature vector and the growth state deep layer feature vector to obtain the corrected feature vector by using a fusion optimization formula as follows: wherein V1 is the growth state shallow layer feature vector, V2 is the growth state deep layer feature vector, <<s and >>s respectively represent left shift and right shift of a feature vector by s bits, and round is a rounding function.

[0048]

[0049]

[0050] wherein V1 is the growth state shallow layer feature vector, V2 is the growth state deep layer feature vector, <<s and >>s respectively represent left shift and right shift of a feature vector by s bits, and round is a rounding function. is the average of all feature values of the growth state shallow layer feature vector V1 and the growth state deep layer feature vector V2, ||·||1 represents a one-norm of a feature vector, d(V1, V2) is the distance between the growth state shallow layer feature vector V1 and the growth state deep layer feature vector V2, and log is a logarithmic function with base 2, and respectively represent position-wise addition and position-wise subtraction, and a and β are weighted hyperparameters, and V' is the corrected feature vector.

[0051] Here, in view of the floating point distribution error and information loss in the vector dimension during the forward propagation process of the growth state shallow layer feature vector V1 and the growth state deep layer feature vector V2 in the network model due to convolution and activation operations, a bit-by-bit shift operation of the vectors is introduced from the perspective of uniform information to balance and standardize the quantization error and information loss during the forward propagation process, and distribution diversity is introduced by reshaping the distribution of feature parameters before fusion, so as to retain information in the form of expanding information entropy. In this way, after linear interpolation is performed on the corrected feature vector V ′ to convert to the same length as the classification feature vector, dot multiplication is performed on the classification feature vector to reduce the information loss of the classification feature vector when classified by the classifier, thereby improving the accuracy of the classification result obtained by the classifier.

[0052] In summary, the crop planting management method based on the embodiments of the present disclosure can use a camera to capture growth state images of the monitored crops and extract growth state feature information about the monitored crops therefrom, so as to intelligently recommend pesticides suitable for the current growth stage, thereby improving the resistance of the crops and agricultural income.

[0053] Figure 6 A block diagram of a crop planting management system 100 according to an embodiment of the present disclosure is shown. As shown, the crop planting management system 100 according to an embodiment of the present disclosure includes an image acquisition module 110 configured to acquire a growth state image of a monitored crop captured by a camera, an image feature extraction module 120 configured to perform image feature extraction on the growth state image to obtain a growth state shallow feature vector and a growth state deep feature vector, and a pesticide type recommendation module 130 configured to determine a recommended pesticide type based on the growth state shallow feature vector and the growth state deep feature vector. Figure 6

[0054] Here, those skilled in the art can understand that the specific functions and operations of each unit and module in the above-described crop planting management system 100 have been described in detail above with reference to the description of the crop planting management method according to an embodiment of the present disclosure, and thus repetitive descriptions thereof will be omitted. Figures 1 to 5

[0055] As described above, the crop planting management system 100 according to an embodiment of the present disclosure can be implemented in various wireless terminals, such as a server having a crop planting management algorithm, etc. In one possible implementation, the crop planting management system 100 according to an embodiment of the present disclosure can be integrated into a wireless terminal as a software module and / or a hardware module. For example, the crop planting management system 100 can be a software module in an operating system of the wireless terminal, or can be an application program developed for the wireless terminal; of course, the crop planting management system 100 can also be one of many hardware modules of the wireless terminal.

[0056] Alternatively, in another example, the crop planting management system 100 and the wireless terminal can also be separate devices, and the crop planting management system 100 can be connected to the wireless terminal through a wired and / or wireless network and transmit interactive information in a predetermined data format.

[0057] Figure 7 An application scenario diagram of a crop planting management method according to an embodiment of the present disclosure is shown. As shown, in the application scenario, first, a growth state image of a monitored crop captured by a camera (for example, C shown in FIG. 1) is acquired, and then the growth state image is input into a server (for example, D shown in FIG. 1) in which a crop planting management algorithm is deployed. Figure 7 Figure 7 Figure 7 Figure 7 ​​​​​The server can process the growth status image using a planting management algorithm of the crop to obtain a classification result for representing a type of pesticide recommended to be used.

[0058] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0059] Embodiments of the present disclosure have been described above, with the understanding that these embodiments are exemplary only, and are not exhaustive of all embodiments of the present disclosure. Many modifications and variations of the described embodiments are possible, in light of the above teachings. It is, therefore, to be understood that changes can be made in the particular embodiments of the technology recited above, those changes being encompassed within the scope of the disclosed technology as defined by the appended claims. The language used in the specification should not be used to argue that what is not specifically defined should be excluded from the technology as defined by the claims. The scope of the technology is defined solely by the claims and the meaning of the terms within the claims.

Claims

1. A method for the cultivation management of a crop, characterized by, The method comprises the following steps: acquiring a growth state image of a monitored crop captured by a camera; extracting image features from the growth state image to obtain a growth state shallow feature vector and a growth state deep feature vector; fusing the growth state shallow feature vector and the growth state deep feature vector using a cascade function to obtain a classification feature vector; forward propagation information preserving fusion is performed on the growth state shallow feature vector and the growth state deep feature vector to obtain a corrected feature vector: forward propagation information preserving fusion is performed on the growth state shallow feature vector and the growth state deep feature vector using the following fusion optimization formula to obtain the corrected feature vector: wherein the fusion optimization formula is: ； ； wherein, is the shallow feature vector of the growth status, is the deep feature vector of the growth status, and respectively represent left shift of the feature vector by bits and right shift by bits, is the rounding function, is the mean of all feature values of the shallow feature vector of the growth status and the deep feature vector of the growth status , represents a norm of the feature vector, is the distance between the shallow feature vector of the growth status and the deep feature vector of the growth status , is the logarithm function with base 2, and respectively represent the positional addition and the positional subtraction, and are the weighted hyperparameters, is the correction feature vector; linear interpolation is performed on the corrected feature vector to obtain an adjusted corrected feature vector, wherein the adjusted corrected feature vector has the same length as the classification feature vector; point multiplication and weighting are performed on the adjusted corrected feature vector and the classification feature vector to obtain an optimized classification feature vector; the optimized classification feature vector is input into a classifier to obtain a classification result, which is used to represent a recommended pesticide type label.

2. The crop planting management method according to claim 1, characterized by, extracting image features from the growth state image to obtain a growth state shallow feature vector and a growth state deep feature vector, comprising: performing image preprocessing on the growth state image to obtain a denoised growth state image; extracting a growth state shallow feature map from the denoised growth state image; extracting a growth state deep feature map from the growth state shallow feature map; respectively performing dimension reduction processing on the growth state shallow feature map and the growth state deep feature map to obtain the growth state shallow feature vector and the growth state deep feature vector.

3. The crop planting management method according to claim 2, characterized by, performing image preprocessing on the growth state image to obtain a denoised growth state image, comprising: performing image denoising processing on the growth state image to obtain the denoised growth state image.

4. The crop planting management method according to claim 3, characterized by, extracting a growth state shallow feature map from the denoised growth state image, comprising: inputting the denoised growth state image into a shallow feature extractor based on a first convolutional neural network model to obtain the growth state shallow feature map.

5. The crop planting management method according to claim 4, characterized by, extracting a growth state deep feature map from the growth state shallow feature map, comprising: inputting the growth state shallow feature map into a deep feature extractor based on a second convolutional neural network model to obtain the growth state deep feature map.

6. The crop planting management method according to claim 5, characterized by, respectively performing dimension reduction processing on the growth state shallow feature map and the growth state deep feature map to obtain the growth state shallow feature vector and the growth state deep feature vector, comprising: respectively unfolding the growth state shallow feature map and the growth state deep feature map into the growth state shallow feature vector and the growth state deep feature vector.

7. The crop planting management method according to claim 6, characterized by, fusing the growth state shallow feature vector and the growth state deep feature vector using a cascade function to obtain a classification feature vector, comprising: fusing the growth state shallow feature vector and the growth state deep feature vector using the following cascade function to obtain the classification feature vector; In the formula, the cascade function is: ； wherein, and both represent point convolution on input, is an activation function, and [] represents a splicing operation, is a feature value of each position in the growth state shallow feature vector, is a feature value of each position in the growth state deep feature vector.

Images