Method and system for judging cadmium tolerance of plant based on multispectrum

Abstract

The invention relates to a method and system for judging cadmium resistance of plants based on multispectrum, and the method comprises the steps: carrying out the image collection of a plant sample through multispectral imaging equipment, and obtaining multispectral original image data; performing image preprocessing on the multispectral original image data to obtain plant multispectral feature data; training a convolutional neural network model through the plant multispectral feature data to obtain a cadmium tolerance judgment model; and performing inference analysis on a multispectral image of a plant to be detected through the cadmium tolerance judgment model to obtain a plant cadmium tolerance judgment result. According to the method, the judgment efficiency can be improved on the premise that destructive sampling is not carried out.

Description

Technical Field

[0001] The invention relates to the technical field of agricultural crop research, and in particular to a method and system for judging plant cadmium tolerance based on multi-spectrum. Background Art

[0002] The determination of plant cadmium tolerance is of great significance for agricultural production, environmental protection and food safety. Traditionally, the determination of whether a plant has cadmium tolerance mainly relies on chemical analysis methods, such as atomic absorption spectroscopy and inductively coupled plasma mass spectrometry. These methods usually require a series of pretreatment steps such as collecting, drying, grinding, and digesting plant samples. Subsequently, the cadmium content in plant tissues is determined by professional chemical instruments, and the cadmium tolerance is determined based on the distribution and accumulation of cadmium in the plant body. At the same time, some researchers have used physiological and biochemical indicators such as antioxidant enzyme activity and changes in chlorophyll content to indirectly evaluate the response of plants to cadmium stress. In recent years, molecular biological methods such as gene expression analysis have also been used to identify plant cadmium resistance-related genes, providing a genetic basis for cadmium resistance determination.

[0003] However, the above traditional methods have obvious shortcomings. Although the chemical analysis method has high accuracy, it has the disadvantages of complex operation, long time consumption, high cost, and the need for professional equipment and technicians, which makes it difficult to meet the needs of large-scale rapid screening. The determination of physiological and biochemical indicators also requires cumbersome laboratory operations, and the changes in a single indicator often cannot fully reflect the plant's cadmium tolerance. Gene expression analysis requires complex molecular biology experimental techniques, which are more expensive and take longer. In addition, most of these methods are destructive and require the collection and processing of plant samples, which is not conducive to continuous monitoring of the same plant individual.

[0004] Therefore, there is an urgent need in this field to develop an efficient, rapid, non-destructive and relatively low-cost method for determining plant cadmium tolerance for large-scale cadmium-tolerant plant screening and environmental monitoring. Summary of the invention

[0005] The invention provides a method and system for judging plant cadmium tolerance based on multi-spectrum, so as to solve the defects of the prior art.

[0006] The first aspect of the present invention provides a method for determining plant cadmium tolerance based on multispectral, comprising:

[0007] S1: Collect images of plant samples through multispectral imaging equipment to obtain multispectral original image data;

[0008] S2: performing image preprocessing on the multispectral original image data to obtain plant multispectral feature data;

[0009] S3: training a convolutional neural network model using the plant multispectral feature data to obtain a cadmium tolerance determination model;

[0010] S4: performing reasoning analysis on the multispectral image of the plant to be tested by using the cadmium tolerance determination model to obtain a plant cadmium tolerance determination result.

[0011] According to a multi-spectral based plant cadmium tolerance determination method provided by the present invention, step S1 further comprises:

[0012] S11: Perform multispectral imaging on the surface of plant leaves according to a preset acquisition angle to obtain multispectral data including visible light band and near infrared band;

[0013] S12: performing spectral decomposition on the multispectral data by using wavelength separation technology to obtain spectral information of different bands;

[0014] S13: performing reflectance correction on the spectral information based on a spectral calibration plate to obtain reflectance-corrected multi-spectral original image data.

[0015] According to a multi-spectral based plant cadmium tolerance determination method provided by the present invention, step S2 specifically includes:

[0016] Using a size adjustment algorithm to unify the size of the multispectral raw image data to generate image data of a fixed size;

[0017] Using a random cropping method to extract local features from the fixed-size image data to obtain a cropped multispectral image;

[0018] Based on the data enhancement technology, the cropped multispectral image is randomly horizontally flipped to obtain an enhanced multispectral image;

[0019] The enhanced multispectral image is standardized according to a preset mean and standard deviation to obtain standardized plant multispectral feature data.

[0020] According to a multi-spectral based plant cadmium tolerance determination method provided by the present invention, step S3 further comprises:

[0021] S31: extracting features from the standardized plant multispectral feature data through a multi-layer convolutional network structure to construct a convolutional neural network model;

[0022] S32: Calculate the loss of the convolutional neural network model based on the cross entropy loss function to obtain the model training loss;

[0023] S33: Based on an adaptive optimization algorithm, the parameters of the convolutional neural network model are optimized and adjusted according to the model training loss to generate an optimized network model;

[0024] S34: Evaluate the performance of the optimized network model based on the validation set accuracy index, and select the cadmium tolerance determination model with the best performance.

[0025] According to a multi-spectral based plant cadmium tolerance determination method provided by the present invention, step S31 further comprises:

[0026] S311: using the first convolutional layer to perform primary feature extraction on the standardized plant multispectral feature data to obtain a first feature map;

[0027] S312: downsampling the first feature map by using a first pooling layer to obtain a second feature map with a reduced size;

[0028] S313: extracting intermediate features from the second feature map through a second convolutional layer to obtain a third feature map;

[0029] S314: further downsampling the third feature map by using a second pooling layer to generate a quadratically reduced fourth feature map;

[0030] S315: performing high-level feature extraction on the fourth feature map according to the third convolutional layer to obtain a fifth feature map;

[0031] S316: Performing a final downsampling on the fifth feature map based on the third pooling layer to obtain a final feature map;

[0032] S317: Flatten the final feature map into a one-dimensional feature vector, and map it to a 128-dimensional potential feature space through a fully connected layer;

[0033] S318: Use the output layer to map the 128-dimensional potential feature space to multiple category output nodes to build a complete convolutional neural network model.

[0034] According to a multi-spectral based plant cadmium tolerance determination method provided by the present invention, step S4 further comprises:

[0035] S41: performing standardization processing on the multispectral image of the plant to be tested through a preprocessing process to obtain standardized feature data of the sample to be tested;

[0036] S42: using the cadmium tolerance determination model to perform forward propagation calculation on the standardized characteristic data of the sample to be tested to obtain an original output logic value;

[0037] S43: Performing probability conversion on the original output logic value based on a Softmax activation function to obtain a category probability distribution;

[0038] S44: determining a category corresponding to a maximum probability according to the category probability distribution, and determining a cadmium tolerance result of the plant to be tested, wherein the cadmium tolerance result includes a cadmium-resistant property and a non-cadmium-resistant property.

[0039] According to a multi-spectral based plant cadmium tolerance determination method provided by the present invention, step S4 further includes:

[0040] A cadmium tolerance confidence index is calculated based on the category probability distribution. When the cadmium tolerance confidence index is lower than a preset threshold, secondary sampling and determination are performed on the plants to be tested to improve the reliability of the determination.

[0041] The second aspect of the present invention provides a plant cadmium tolerance judgment system based on multispectral, comprising:

[0042] An acquisition module, wherein the acquisition module is configured as a multispectral imaging device, and is used to acquire images of plant samples to obtain multispectral original image data;

[0043] A preprocessing module, the preprocessing module is used to perform image preprocessing on the multispectral original image data to obtain plant multispectral feature data;

[0044] A training module, wherein the training module is used to train a convolutional neural network model using the plant multispectral feature data to obtain a cadmium tolerance determination model;

[0045] The determination module is used to perform reasoning analysis on the multispectral image of the plant to be tested through the cadmium tolerance determination model trained by the training module to obtain the plant cadmium tolerance determination result.

[0046] The third aspect of the present invention provides a multi-spectral based plant cadmium tolerance judgment device, comprising:

[0047] A memory and at least one processor, wherein instructions are stored in the memory;

[0048] At least one of the processors calls the instructions in the memory to enable a multi-spectrum-based plant cadmium tolerance judgment device to execute a multi-spectrum-based plant cadmium tolerance judgment method as described in any one of the above.

[0049] A fourth aspect of the present invention provides a computer-readable storage medium having instructions stored thereon, and when the instructions are executed by a processor, a multi-spectral-based method for determining plant cadmium tolerance as described in any one of the above items is implemented.

[0050] The present invention provides a multi-spectral based plant cadmium tolerance judgment method, system, device and storage medium. The multi-spectral imaging device is used to perform non-destructive imaging of plants, and a cadmium tolerance judgment model is established in combination with a deep learning algorithm. This not only changes the limitation of traditional chemical analysis methods that require destructive sampling, but also greatly improves the judgment efficiency and realizes large-scale plant cadmium tolerance screening. In the present invention, the specific structural design of the convolutional neural network algorithm, especially the design of the multi-layer convolution structure, enables the model to gradually extract key features related to cadmium stress response in plant multi-spectral images from shallow to deep layers. The hierarchical feature extraction capability enables the model to automatically learn subtle changes in plant spectral characteristics under cadmium stress without manually designing feature extraction rules. At the same time, the cross entropy loss function and adaptive optimization algorithm used in the model training process enable the model to fit the training data more accurately, and the design of the fully connected layer realizes effective mapping from high-dimensional feature space to low-dimensional judgment space.

[0051] In the application of the present invention, no professional chemical knowledge and complex instruments and equipment are required. It is only necessary to collect plant images through a multispectral camera, and the determination result can be quickly obtained through a preset processing flow, which greatly reduces the technical threshold and operating cost. In particular, in the field of agricultural breeding, a large number of cadmium-resistant plant varieties can be quickly screened out, providing efficient tools for the breeding of heavy metal-resistant crops. In the field of environmental remediation, plant species suitable for cadmium-contaminated soil remediation can be quickly identified, accelerating the promotion and application of plant remediation technology. In the field of food safety, it can be used for early warning of cadmium resistance of crops in agricultural product production bases, and effectively preventing and controlling cadmium-contaminated agricultural products from entering the food chain. In general, the method of the present invention has significant advantages such as non-destructive, rapid, accurate, and low cost, and provides innovative technical means for the research and application of plant cadmium resistance, and has broad application prospects and practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0053] Figure 1 A schematic diagram of a multi-spectral method for determining plant cadmium tolerance provided by the present invention;

[0054] Figure 2 A schematic diagram of the structure of a plant cadmium tolerance judgment system based on multispectrum provided by the present invention.

[0055] Figure numerals: 100, acquisition module; 200, pre-processing module; 300, training module; 400, determination module. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments, and they should not be understood as limitations on the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In the description of the present invention, it should be understood that the terms used are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0057] The embodiments of the present invention are described below with reference to the accompanying drawings.

[0058] like Figure 1 As shown, the present invention provides a method for judging plant cadmium tolerance based on multispectral, comprising:

[0059] S1: Capture images of plant samples using a multispectral imaging device to obtain multispectral raw image data.

[0060] Furthermore, in step S1, the plant sample is first imaged by a multispectral imaging device to obtain multispectral raw image data. A multispectral imaging device is a special device that can simultaneously collect the spectral reflectance of a target object in multiple different bands. Its working principle is to use a spectroscopic system to decompose the incident light into multiple bands, and record the intensity information of each band through a sensor. In the present invention, the multispectral imaging device used is equipped with a special optical filter group that can collect spectral information from visible light to near-infrared bands. These filters generally cover multiple discrete bands within the wavelength range of 400-1000nm. Multispectral imaging is different from ordinary RGB imaging. It can record the reflection characteristics of the target object at multiple narrowband wavelengths. These characteristics are often closely related to the physiological and biochemical state of the plant, and therefore provide basic data for non-destructive judgment of plant cadmium resistance.

[0061] Wherein, step S1 further comprises:

[0062] S11: Perform multispectral imaging on the surface of plant leaves according to a preset acquisition angle to obtain multispectral data including visible light band and near infrared band.

[0063] Furthermore, multispectral imaging of the plant leaf surface at a preset acquisition angle is a key step in obtaining high-quality multispectral data. The preset acquisition angle is usually selected to be perpendicular to the leaf surface or at a 45-degree angle to the leaf surface. This angle setting helps to reduce signal interference caused by light reflection while ensuring image clarity.

[0064] S12: performing spectral decomposition on the multi-spectral data by using wavelength separation technology to obtain spectral information of different bands.

[0065] Furthermore, wavelength separation technology refers to a method of separating mixed wavelength light signals into single wavelength or narrow band light signals by optical or algorithmic means. In the present invention, a wavelength separation technology based on a filter array is adopted, and each filter only allows light in a specific wavelength range to pass, thereby realizing the separation of light signals in different bands. In specific operation, the built-in filter array of the multi-spectral imaging device will automatically complete the wavelength separation process, decompose the reflected light into multiple preset bands, and each band corresponds to a sensor unit for signal acquisition. Through step S12, the spectral information of multiple bands can be obtained, and each band represents the reflection characteristics of the plant at a specific wavelength.

[0066] S13: performing reflectance correction on the spectral information based on a spectral calibration plate to obtain reflectance-corrected multi-spectral original image data.

[0067] Furthermore, the spectral calibration plate is a standard reference object with a known reflectivity characteristic on the surface, which usually contains multiple areas with different reflectivity, from a black area with a reflectivity close to 0% to a white area with a reflectivity close to 100%. In the multispectral imaging process, the spectral calibration plate is first imaged to record its original grayscale values ​​in each band, and then a mapping relationship is established between the known reflectivity data of the calibration plate and the original grayscale value to generate a reflectivity conversion function. Subsequently, this conversion function is applied to the original grayscale image of the plant sample to obtain the reflectivity-corrected multispectral original image data. The core of reflectivity correction is to eliminate the influence of environmental factors such as lighting conditions and atmospheric conditions on the imaging results, so that images collected at different times and under different conditions are comparable. In the specific calculation process, for each band of the image, a linear or nonlinear mapping relationship is first established using the known reflectivity of the calibration plate and its corresponding grayscale value, and then this relationship is applied to each pixel of the plant sample image, thereby converting the original grayscale value into an actual reflectivity value.

[0068] S2: performing image preprocessing on the multispectral original image data to obtain plant multispectral feature data.

[0069] Wherein, step S2 specifically includes:

[0070] The multi-spectral original image data is processed to uniform size using a size adjustment algorithm to generate image data of a fixed size.

[0071] Furthermore, the resizing algorithm refers to a technique for converting images of different sizes into a uniform size by interpolation or sampling methods. In the present invention, a bilinear interpolation algorithm is used, which determines the new pixel value by calculating the weighted average of the original pixels in a 2×2 neighborhood around the target pixel position, thereby ensuring image quality and controlling computational complexity.

[0072] A random cropping method is used to extract local features from the fixed-size image data to obtain a cropped multispectral image.

[0073] Furthermore, random cropping is a technique for data enhancement, which increases data diversity and prevents model overfitting by randomly selecting a part of the original image as training samples.

[0074] The cropped multispectral image is randomly horizontally flipped based on data enhancement technology to obtain an enhanced multispectral image.

[0075] Random horizontal flipping of the cropped multispectral image based on data augmentation technology is a means to further expand the data set. Data augmentation refers to the technology of generating new training samples by performing a series of transformations on the original data. The purpose is to increase data diversity and improve the generalization ability of the model. Random horizontal flipping is to mirror the image along the vertical axis so that the left and right sides are swapped.

[0076] The enhanced multispectral image is standardized according to a preset mean and standard deviation to obtain standardized plant multispectral feature data.

[0077] Furthermore, standardization is the process of converting data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The purpose is to eliminate the dimensional differences between different features and accelerate the convergence of the model. In the present invention, standardization is performed on the images of each band separately, using the preset mean [0.485, 0.456, 0.406] and standard deviation [0.229, 0.224, 0.225]. This set of values ​​is based on empirical values ​​obtained from statistics of a large number of plant multispectral images and is applicable to most plant samples.

[0078] S3: Training a convolutional neural network model using the plant multispectral feature data to obtain a cadmium tolerance determination model.

[0079] In step S3 of the present invention, feature extraction is performed on standardized plant multispectral feature data through a multi-layer convolutional network structure. Constructing a convolutional neural network model is the core link for realizing automatic judgment of plant cadmium resistance. A convolutional neural network (CNN) is a deep learning model specially used for processing data with a grid structure, and is particularly suitable for image data processing. The multi-layer convolutional network structure consists of multiple convolutional layers, pooling layers and fully connected layers, and can automatically learn the hierarchical feature representation of data, from low-level features to high-level features, and extract and abstract key information in the image layer by layer.

[0080] Wherein, step S3 further comprises:

[0081] S31: extracting features from the standardized plant multispectral feature data through a multi-layer convolutional network structure, and constructing a convolutional neural network model.

[0082] Wherein, step S31 further comprises:

[0083] S311: Use the first convolution layer to perform primary feature extraction on the standardized plant multispectral feature data to obtain a first feature map; S312: Use the first pooling layer to downsample the first feature map to obtain a second feature map of reduced size; S313: Perform intermediate feature extraction on the second feature map through the second convolution layer to obtain a third feature map; S314: Use the second pooling layer to further downsample the third feature map to generate a quadratically reduced fourth feature map; S315: Perform high-level feature extraction on the fourth feature map based on the third convolution layer to obtain a fifth feature map; S316: Perform final downsampling on the fifth feature map based on the third pooling layer to obtain a final feature map; S317: Flatten the final feature map into a one-dimensional feature vector, and map it to a 128-dimensional potential feature space through a fully connected layer; S318: Use the output layer to map the 128-dimensional potential feature space to multiple category output nodes to build a complete convolutional neural network model.

[0084] In a specific embodiment, the feature extraction process of the multi-layer convolutional network is illustrated by taking the cadmium tolerance judgment of rice leaves as an example. First, the standardized rice leaf multispectral image data (224×224×16, assuming there are 16 spectral bands) obtained after preprocessing is input into the first convolution layer, and the 16 3×3 convolution kernels of the first convolution layer perform convolution operations on the input data respectively to generate 16 feature maps. Taking one of the convolution kernels as an example, it is specifically sensitive to the reflectivity changes of the chlorophyll absorption band (red light band). When sliding to the central area of ​​the leaf, if the standardized reflectivity value of the area in the red light band is low (such as -1.5, indicating that it is lower than the average level), and higher in the near-infrared band (such as 1.8, indicating that it is higher than the average level), then this typical plant spectral feature will activate the convolution kernel and produce a higher output value (such as 2.3). The 16 convolution kernels capture different primary features respectively to form 16 222×222 feature maps. These feature maps are processed by the 2×2 maximum pooling of the first pooling layer, and the size is reduced to 111×111×16. Next, the 32 3×3 convolution kernels of the second convolution layer perform convolution operations on these 16 feature maps. Each convolution kernel convolves all 16 feature maps and adds the results, plus the bias term, and outputs a feature map after passing through the ReLU activation function. For example, if one of the convolution kernels specifically recognizes the vein area, it will give high weight to the primary features containing linear structures, and produce a high activation value when sliding to the vein position. The 32 convolution kernels extract different intermediate features respectively, forming 32 109×109 feature maps. These feature maps are processed by the second pooling layer and the size is reduced to 54×54×32. Then, the 64 3×3 convolution kernels of the third convolution layer perform convolution operations on these 32 feature maps to extract higher-level features. At this time, the convolution kernel may specifically recognize the spectral patterns unique to cadmium stress, such as the movement of the red edge position or the reflectance ratio of a specific band. The 64 convolution kernels extract different high-level features to form 64 52×52 feature maps. These feature maps are processed by the third pooling layer and reduced in size to 26×26×64 to form the final feature map. The final feature map contains the most critical feature information in the original multispectral image, especially the features related to rice cadmium resistance. Next, the final feature map is flattened into a one-dimensional vector and mapped to a 128-dimensional latent feature space through a fully connected layer. These 128 nodes represent the core features required to judge rice cadmium resistance, such as the reflectance pattern of a specific band, the ratio relationship between bands, etc. Finally, the output layer maps the 128-dimensional features to 2 output nodes, which represent the possibility that the rice sample belongs to a "cadmium-resistant variety" and a "non-cadmium-resistant variety" respectively. Through this multi-level feature extraction and transformation, the CNN model can automatically learn the key features required to judge the cadmium resistance of rice leaves from the multispectral images of rice leaves, and achieve fast and accurate cadmium resistance judgment.

[0085] S32: Perform loss calculation on the convolutional neural network model based on the cross entropy loss function to obtain the model training loss.

[0086] The cross entropy loss function used in the present invention is a loss function in the classification task, which can measure the degree of difference between the predicted distribution and the true distribution. By averaging the losses of all training samples, the overall training loss of the model is obtained. This loss value directly reflects the classification performance under the current model parameters.

[0087] S33: Based on an adaptive optimization algorithm, the parameters of the convolutional neural network model are optimized and adjusted according to the model training loss to generate an optimized network model.

[0088] Furthermore, the adaptive optimization algorithm is a type of optimization method that can automatically adjust the learning rate to adapt to the learning requirements of different parameters. In the present invention, the Adam optimization algorithm is adopted. The Adam optimization algorithm combines the advantages of the momentum method and RMSprop and can effectively handle sparse gradients and noise.

[0089] S34: Evaluate the performance of the optimized network model based on the validation set accuracy index, and select the cadmium tolerance determination model with the best performance.

[0090] Furthermore, the validation set accuracy refers to the correct prediction ratio of the model on the validation data that did not participate in the training, and the calculation formula is: Accuracy = Number of correctly predicted samples / Total number of samples. In step S34 of the present invention, after each training cycle (epoch), the model is applied to the validation set to calculate the accuracy. This model selection strategy based on validation set performance ensures that the final model has the best generalization ability and can accurately judge the cadmium resistance of unknown plant samples.

[0091] S4: performing reasoning analysis on the multispectral image of the plant to be tested by using the cadmium tolerance determination model to obtain a plant cadmium tolerance determination result.

[0092] Wherein, step S4 further comprises:

[0093] S41: Standardizing the multispectral image of the plant to be tested through a preprocessing process to obtain standardized feature data of the sample to be tested.

[0094] The standardization process in step S41 adopts the same preprocessing process as that in the training stage, including operations such as resizing, cropping, and standardization.

[0095] S42: Perform forward propagation calculation on the standardized characteristic data of the sample to be tested using the cadmium tolerance determination model to obtain an original output logic value.

[0096] Furthermore, the forward propagation calculation in step S42 refers to the calculation process of inputting data from the input layer through each hidden layer to the output layer in accordance with the network structure of the model. In the present invention, the standardized feature data is first input into the first convolutional layer, and then through a series of operations such as convolution operation, activation function, pooling, etc., and then through the second convolutional layer, the third convolutional layer and its corresponding pooling layer in turn, and finally outputs the original logical value through the fully connected layer. The entire calculation process is completely carried out according to the network structure during model training, but does not involve back propagation and parameter update.

[0097] S43: Perform probability conversion on the original output logic value based on the Softmax activation function to obtain category probability distribution.

[0098] Furthermore, the Softmax function is a function that converts any real number vector into a probability distribution, ensuring that all output values ​​are between 0 and 1 and the sum is 1. For example, for the original output logical value [2.5, -1.8], the exponential value e^2.5≈12.18 and e^(-1.8)≈0.165 are first calculated, then the sum 12.18+0.165≈12.345 is calculated, and finally the probability values ​​p1=12.18 / 12.345≈0.987 and p2=0.165 / 12.345≈0.013 are calculated. The probability distribution obtained after conversion is [0.987, 0.013], indicating that the sample has a 98.7% probability of being a "cadmium-tolerant plant" and a 1.3% probability of being a "cadmium-intolerant plant". This probability output not only gives the classification result, but also provides confidence information for the result.

[0099] S44: determining a category corresponding to a maximum probability according to the category probability distribution, and determining a cadmium tolerance result of the plant to be tested, wherein the cadmium tolerance result includes a cadmium-resistant property and a non-cadmium-resistant property.

[0100] In step S4, finally, determining the category corresponding to the maximum probability based on the category probability distribution is a simple and effective method to obtain the final judgment result. After obtaining the probability distribution, the category with the largest probability value is selected as the final judgment result. For example, for the probability distribution [0.987, 0.013], the probability value of the first category "cadmium-resistant plants" is 0.987, which is the largest, so the plant sample is judged to have cadmium resistance. Specifically, if the index corresponding to the maximum probability value is 0, the judgment result is "having cadmium resistance", and if the corresponding index is 1, the judgment result is "not having cadmium resistance". This judgment method based on the maximum probability is intuitive and effective, and can give a clear binary classification result.

[0101] Wherein, step S4 also includes:

[0102] A cadmium tolerance confidence index is calculated based on the category probability distribution. When the cadmium tolerance confidence index is lower than a preset threshold, secondary sampling and determination are performed on the plants to be tested to improve the reliability of the determination.

[0103] In addition, calculating the cadmium tolerance confidence index based on the category probability distribution and making a secondary judgment is an important supplementary mechanism to improve the reliability of the results. The cadmium tolerance confidence index usually directly uses the maximum probability value as the main indicator, or calculates the difference between the maximum probability and the second largest probability as the judgment difference.

[0104] For example, if the probability distribution of a sample is [0.75, 0.25], and the confidence index is 0.75 or 0.5, which is lower than the preset threshold of 0.8, it is necessary to re-collect the multispectral image of the plant, possibly at different parts or at different growth stages, and then repeat the above reasoning process. Finally, a more reliable cadmium tolerance judgment is obtained by combining multiple judgment results. The confidence-based secondary judgment mechanism effectively improves the reliability of the judgment results and is particularly suitable for processing edge cases.

[0105] like Figure 2 As shown, the present invention also provides a plant cadmium tolerance judgment system based on multi-spectrum, comprising:

[0106] A collection module 100, which is configured as a multispectral imaging device, is used to collect images of plant samples and obtain multispectral original image data;

[0107] A preprocessing module 200, the preprocessing module 200 is used to perform image preprocessing on the multispectral original image data to obtain plant multispectral feature data;

[0108] A training module 300, wherein the training module 300 is used to train a convolutional neural network model using the plant multispectral feature data to obtain a cadmium tolerance determination model;

[0109] The determination module 400 is used to perform reasoning analysis on the multispectral image of the plant to be tested by using the cadmium tolerance determination model trained by the training module 300 to obtain a plant cadmium tolerance determination result.

[0110] The present invention also provides a multi-spectral based plant cadmium tolerance judgment device, comprising:

[0111] A memory and at least one processor, wherein instructions are stored in the memory;

[0112] At least one of the processors calls the instructions in the memory to enable a multi-spectrum-based plant cadmium tolerance judgment device to execute a multi-spectrum-based plant cadmium tolerance judgment method as described in any one of the above.

[0113] The present invention also provides a computer-readable storage medium having instructions stored thereon, and when the instructions are executed by a processor, a multi-spectral-based method for determining plant cadmium tolerance as described in any one of the above is implemented.

[0114] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0115] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for determining plant cadmium tolerance based on multispectral analysis, characterized in that: include: S1: Collect images of plant samples through multispectral imaging equipment to obtain multispectral original image data; S2: performing image preprocessing on the multispectral original image data to obtain plant multispectral feature data; S3: training a convolutional neural network model using the plant multispectral feature data to obtain a cadmium tolerance determination model; S4: performing reasoning analysis on the multispectral image of the plant to be tested by using the cadmium tolerance determination model to obtain a plant cadmium tolerance determination result.

2. A method for determining plant cadmium tolerance based on multispectrum according to claim 1, characterized in that: Step S1 further comprises: S11: Perform multispectral imaging on the surface of plant leaves according to a preset acquisition angle to obtain multispectral data including visible light band and near infrared band; S12: performing spectral decomposition on the multispectral data by using wavelength separation technology to obtain spectral information of different bands; S13: performing reflectivity correction on the spectral information based on a spectral calibration plate to obtain reflectivity-corrected multi-spectral original image data.

3. A method for determining plant cadmium tolerance based on multispectrum according to claim 1, characterized in that: Step S2 specifically includes: Using a size adjustment algorithm to unify the size of the multispectral raw image data to generate image data of a fixed size; Using a random cropping method to extract local features from the fixed-size image data to obtain a cropped multispectral image; Based on the data enhancement technology, the cropped multispectral image is randomly horizontally flipped to obtain an enhanced multispectral image; The enhanced multispectral image is standardized according to a preset mean and standard deviation to obtain standardized plant multispectral feature data.

4. A method for determining plant cadmium tolerance based on multispectrum according to claim 1, characterized in that: Step S3 further comprises: S31: extracting features from the standardized plant multispectral feature data through a multi-layer convolutional network structure to construct a convolutional neural network model; S32: Calculate the loss of the convolutional neural network model based on the cross entropy loss function to obtain the model training loss; S33: Based on an adaptive optimization algorithm, the parameters of the convolutional neural network model are optimized and adjusted according to the model training loss to generate an optimized network model; S34: Evaluate the performance of the optimized network model based on the validation set accuracy index, and select the cadmium tolerance determination model with the best performance.

5. A method for determining plant cadmium tolerance based on multispectrum according to claim 4, characterized in that: Step S31 further includes: S311: using the first convolutional layer to perform primary feature extraction on the standardized plant multispectral feature data to obtain a first feature map; S312: downsampling the first feature map by using a first pooling layer to obtain a second feature map with a reduced size; S313: extracting intermediate features from the second feature map through a second convolutional layer to obtain a third feature map; S314: further downsampling the third feature map by using a second pooling layer to generate a quadratically reduced fourth feature map; S315: performing high-level feature extraction on the fourth feature map according to the third convolutional layer to obtain a fifth feature map; S316: Performing a final downsampling on the fifth feature map based on the third pooling layer to obtain a final feature map; S317: Flatten the final feature map into a one-dimensional feature vector, and map it to a 128-dimensional potential feature space through a fully connected layer; S318: Use the output layer to map the 128-dimensional potential feature space to multiple category output nodes to build a complete convolutional neural network model.

6. A method for determining plant cadmium tolerance based on multispectrum according to claim 1, characterized in that: Step S4 further comprises: S41: performing standardization processing on the multispectral image of the plant to be tested through a preprocessing process to obtain standardized feature data of the sample to be tested; S42: using the cadmium tolerance determination model to perform forward propagation calculation on the standardized characteristic data of the sample to be tested to obtain an original output logic value; S43: Performing probability conversion on the original output logic value based on a Softmax activation function to obtain a category probability distribution; S44: determining a category corresponding to a maximum probability according to the category probability distribution, and determining a cadmium tolerance result of the plant to be tested, wherein the cadmium tolerance result includes a cadmium-resistant property and a non-cadmium-resistant property.

7. A method for determining plant cadmium tolerance based on multispectrum according to claim 6, characterized in that: Step S4 also includes: A cadmium tolerance confidence index is calculated based on the category probability distribution. When the cadmium tolerance confidence index is lower than a preset threshold, secondary sampling and determination are performed on the plants to be tested to improve the reliability of the determination.

8. A plant cadmium tolerance judgment system based on multispectral, characterized in that: include: An acquisition module, wherein the acquisition module is configured as a multispectral imaging device, and is used to acquire images of plant samples to obtain multispectral original image data; A preprocessing module, the preprocessing module is used to perform image preprocessing on the multispectral original image data to obtain plant multispectral feature data; A training module, wherein the training module is used to train a convolutional neural network model using the plant multispectral feature data to obtain a cadmium tolerance determination model; The determination module is used to perform reasoning analysis on the multispectral image of the plant to be tested through the cadmium tolerance determination model trained by the training module to obtain the plant cadmium tolerance determination result.

9. A multi-spectral based plant cadmium tolerance judgment device, characterized in that: include: A memory and at least one processor, wherein instructions are stored in the memory; At least one of the processors calls the instructions in the memory to enable a multi-spectrum-based plant cadmium tolerance judgment device to execute a multi-spectrum-based plant cadmium tolerance judgment method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed by the processor, a multi-spectrum-based method for determining plant cadmium tolerance is implemented as described in any one of claims 1 to 7.

Images