CN117036088A - Data acquisition and analysis method for identifying growth situation of greening plants by AI - Google Patents
Data acquisition and analysis method for identifying growth situation of greening plants by AI Download PDFInfo
- Publication number
- CN117036088A CN117036088A CN202311051190.5A CN202311051190A CN117036088A CN 117036088 A CN117036088 A CN 117036088A CN 202311051190 A CN202311051190 A CN 202311051190A CN 117036088 A CN117036088 A CN 117036088A
- Authority
- CN
- China
- Prior art keywords
- data
- model
- growth
- plants
- plant
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000012010 growth Effects 0.000 title claims abstract description 143
- 238000004458 analytical method Methods 0.000 title claims abstract description 30
- 241000196324 Embryophyta Species 0.000 claims abstract description 112
- 230000008635 plant growth Effects 0.000 claims abstract description 50
- 238000000034 method Methods 0.000 claims abstract description 49
- 238000012549 training Methods 0.000 claims abstract description 45
- 239000002689 soil Substances 0.000 claims abstract description 40
- 230000003595 spectral effect Effects 0.000 claims abstract description 20
- 230000008569 process Effects 0.000 claims abstract description 17
- 230000000737 periodic effect Effects 0.000 claims abstract description 11
- 238000013528 artificial neural network Methods 0.000 claims abstract description 7
- 230000036541 health Effects 0.000 claims abstract description 7
- 238000012300 Sequence Analysis Methods 0.000 claims abstract description 6
- 230000007613 environmental effect Effects 0.000 claims description 36
- 241000607479 Yersinia pestis Species 0.000 claims description 28
- 238000012545 processing Methods 0.000 claims description 25
- 201000010099 disease Diseases 0.000 claims description 21
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims description 21
- 230000002159 abnormal effect Effects 0.000 claims description 20
- 230000006870 function Effects 0.000 claims description 20
- 238000012795 verification Methods 0.000 claims description 16
- 238000005457 optimization Methods 0.000 claims description 15
- 230000002262 irrigation Effects 0.000 claims description 14
- 238000003973 irrigation Methods 0.000 claims description 14
- 238000005286 illumination Methods 0.000 claims description 12
- 238000004422 calculation algorithm Methods 0.000 claims description 11
- 238000012544 monitoring process Methods 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 10
- 238000004140 cleaning Methods 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 9
- 238000005311 autocorrelation function Methods 0.000 claims description 8
- 230000001965 increasing effect Effects 0.000 claims description 8
- 238000011176 pooling Methods 0.000 claims description 8
- 230000004720 fertilization Effects 0.000 claims description 7
- 230000007774 longterm Effects 0.000 claims description 7
- 238000007726 management method Methods 0.000 claims description 7
- 239000000575 pesticide Substances 0.000 claims description 7
- 238000007637 random forest analysis Methods 0.000 claims description 7
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 6
- 238000013135 deep learning Methods 0.000 claims description 6
- 238000013136 deep learning model Methods 0.000 claims description 6
- 238000005516 engineering process Methods 0.000 claims description 6
- 239000003337 fertilizer Substances 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 6
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 230000001932 seasonal effect Effects 0.000 claims description 6
- 238000012731 temporal analysis Methods 0.000 claims description 6
- 238000000700 time series analysis Methods 0.000 claims description 6
- 230000005856 abnormality Effects 0.000 claims description 5
- 238000013527 convolutional neural network Methods 0.000 claims description 5
- 235000013399 edible fruits Nutrition 0.000 claims description 5
- 238000004088 simulation Methods 0.000 claims description 5
- 238000001228 spectrum Methods 0.000 claims description 5
- 238000003860 storage Methods 0.000 claims description 5
- 238000010200 validation analysis Methods 0.000 claims description 5
- 241000238631 Hexapoda Species 0.000 claims description 4
- 238000003066 decision tree Methods 0.000 claims description 4
- 230000003247 decreasing effect Effects 0.000 claims description 4
- 230000003993 interaction Effects 0.000 claims description 4
- 238000013508 migration Methods 0.000 claims description 4
- 230000005012 migration Effects 0.000 claims description 4
- 238000003062 neural network model Methods 0.000 claims description 4
- 230000002787 reinforcement Effects 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000008901 benefit Effects 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 230000002708 enhancing effect Effects 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 230000010354 integration Effects 0.000 claims description 3
- 238000012417 linear regression Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 241001464837 Viridiplantae Species 0.000 claims description 2
- 238000000354 decomposition reaction Methods 0.000 claims description 2
- 238000011049 filling Methods 0.000 claims description 2
- 238000009499 grossing Methods 0.000 claims description 2
- 230000009545 invasion Effects 0.000 claims description 2
- 238000012423 maintenance Methods 0.000 claims description 2
- 239000011159 matrix material Substances 0.000 claims description 2
- 230000000877 morphologic effect Effects 0.000 claims description 2
- 230000035945 sensitivity Effects 0.000 claims description 2
- 241000208818 Helianthus Species 0.000 description 10
- 235000003222 Helianthus annuus Nutrition 0.000 description 10
- 241001092500 Photinia x fraseri Species 0.000 description 10
- 241001116500 Taxus Species 0.000 description 9
- 230000002354 daily effect Effects 0.000 description 7
- 241001149649 Taxus wallichiana var. chinensis Species 0.000 description 6
- 244000111489 Gardenia augusta Species 0.000 description 5
- 235000018958 Gardenia augusta Nutrition 0.000 description 5
- 240000004144 Acer rubrum Species 0.000 description 4
- 235000011772 Acer rubrum var tomentosum Nutrition 0.000 description 4
- 235000009057 Acer rubrum var tridens Nutrition 0.000 description 4
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 4
- 238000013480 data collection Methods 0.000 description 3
- 230000003203 everyday effect Effects 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 241000488583 Panonychus ulmi Species 0.000 description 2
- 235000008331 Pinus X rigitaeda Nutrition 0.000 description 2
- 235000011613 Pinus brutia Nutrition 0.000 description 2
- 241000018646 Pinus brutia Species 0.000 description 2
- 238000007405 data analysis Methods 0.000 description 2
- 230000001934 delay Effects 0.000 description 2
- 238000003306 harvesting Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 238000010183 spectrum analysis Methods 0.000 description 2
- 241000157835 Gardenia Species 0.000 description 1
- 206010061217 Infestation Diseases 0.000 description 1
- 241001465754 Metazoa Species 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000009418 agronomic effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000000853 biopesticidal effect Effects 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000010219 correlation analysis Methods 0.000 description 1
- 230000000875 corresponding effect Effects 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000000459 effect on growth Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 235000021384 green leafy vegetables Nutrition 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 239000002917 insecticide Substances 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010223 real-time analysis Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004611 spectroscopical analysis Methods 0.000 description 1
- 238000005507 spraying Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000036962 time dependent Effects 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 235000013311 vegetables Nutrition 0.000 description 1
- 238000011179 visual inspection Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
- 230000003442 weekly effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/02—Agriculture; Fishing; Mining
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
Abstract
The invention relates to a data acquisition and analysis method for identifying the growth situation of greening plants by AI. The method comprises the following steps: s1, collecting plant growth data from multiple angles by utilizing a multi-mode sensor comprising an RGB camera, an infrared camera, a temperature and humidity sensor and a soil humidity sensor, and fusing the sensor data by a neural network method; s2, performing time sequence analysis on the obtained continuous plant growth data, identifying a growth curve and a periodic mode, and predicting a future growth situation according to the past data; s3, collecting spectral reflection data of plant leaves by utilizing a spectrometer, and training an AI model based on the spectral data to identify and evaluate different growth states and health conditions of plants; s4, three-dimensional scanning is carried out on the plants by using a camera or a laser radar, three-dimensional data of the plants are obtained, the growth process of the plants is further simulated by using the three-dimensional data, and the growth situation of the plants is estimated by comparing the growth process with actual growth data.
Description
Technical Field
The invention relates to a data acquisition and analysis method for identifying the growth situation of greening plants by AI.
Background
AI identification greening plant growth situation typically combines image recognition technology, spectral analysis and environmental sensor data to comprehensively evaluate plant growth situation. The plant is scanned through the camera and the spectrometer, the data of the form and the health state of the plant are collected, and the growth state, diseases and insect pest risks of the plant can be predicted and estimated by the AI model by combining the air temperature, the humidity, the illumination and the soil condition data collected by the environment sensor.
But this method is severely dependent on the quality of the data. The analysis results may be inaccurate if there is a bias or noise in the collected image, spectrum, or environmental data. In addition, spectrometers and high quality cameras are costly and may not be suitable for large areas of farms or greens. Processing and analyzing large amounts of data requires high performance computing power and may lead to delays, especially in real-time analysis. Although environmental sensors can collect a variety of data, certain micro-environmental factors that have a significant impact on plant growth may be missed. Moreover, AI models trained on plants of one region or class may not be suitable for other regions or classes because there may be differences in growth conditions and plant responses. For situations where real-time intervention is required, delays in data collection, processing, and feedback may result in missing the best opportunity for intervention. In addition, the cameras and sensors may be disturbed by weather, animals or other external factors, affecting the quality of the data. Moreover, plant growth and the disease or pest it is exposed to may change over time, which requires periodic updating of the AI model, which may otherwise lead to prediction bias.
Thus, although the method of AI to identify the growth situation of greening plants has great potential in theory, there are still many challenges and limitations in practical application. This requires further technical innovation and optimization to address.
Disclosure of Invention
The invention aims to provide a data acquisition and analysis method for identifying the growth situation of greening plants by using an AI, so that the defects and shortcomings pointed out in the background art are overcome.
The invention solves the technical problems as follows: the method comprises the following steps:
s1, collecting plant growth data from multiple angles by utilizing a multi-mode sensor comprising an RGB camera, an infrared camera, a temperature and humidity sensor and a soil humidity sensor, and fusing the sensor data by a neural network method;
s2, performing time sequence analysis on the obtained continuous plant growth data, identifying a growth curve and a periodic mode, and predicting a future growth situation according to the past data;
s3, collecting spectral reflection data of plant leaves by utilizing a spectrometer, and training an AI model based on the spectral data to identify and evaluate different growth states and health conditions of plants;
s4, three-dimensional scanning is carried out on the plants by using a camera or a laser radar, three-dimensional data of the plants are obtained, the growth process of the plants is further simulated by using the three-dimensional data, and the growth situation of the plants is estimated by comparing the growth process with actual growth data;
S5, collecting environmental factor data related to plant growth, including air temperature, humidity, illumination and soil conditions, by integrating an environmental sensor, and predicting growth situations of plants under different environmental conditions by combining the environmental factor data and using an AI model;
s6, applying a deep learning method, and enabling the AI model to automatically identify and evaluate the growth stage, diseases and pest invasion states of plants through training a large number of plant images;
s7, further constructing an intelligent recommendation system on the basis of analyzing plant growth situation by the AI model, and providing targeted planting suggestions for gardeners or farmers.
Further, the multi-mode sensor data processing process comprises the following steps: configuring a set of sensors to systematically collect data of plants and their environments from a plurality of angles and dimensions; preprocessing data from different sources, eliminating noise, and performing standardization and normalization operation; extracting key features related to plant growth situation from data of each mode by using a neural network structure, wherein the key features comprise morphological features of plants extracted from RGB images by using a convolutional neural network or environmental change features extracted from temperature and humidity data by using a time sequence analysis method;
According to the demands of plant growth situations, adopting weighting, stacking or cascading strategies to integrate the characteristics of different modes into a unified characteristic representation; training an AI model by using the fused features, and dynamically adjusting the strategy and weight of data fusion according to the data collected in real time and the evaluation result of the AI model;
the weighting strategy is calculated by the following expression:
F=w 1 ×F 1 +w 2 ×F 2
wherein F is a fused feature, F 1 And F 2 Is characteristic of two modes, w 1 And w 2 Is their weight.
Further, the time series analysis adopts continuous collection of time series data related to plant growth, including factors of plant height, leaf number, leaf area and environment thereof; then, data cleaning is carried out, wherein the data cleaning comprises filling of missing values and abnormal value removal;
then using fourier transforms or seasonal decomposition to identify periodic patterns in the time series; the specific formula is as follows:
wherein X (t) is a time series, a 0 ,a n ,b n Is a coefficient, N is a period, f 0 Is the fundamental frequency;
then, linear regression and moving average technology are used for identifying long-term trend, so that the growth situation of the plant in the future is predicted;
in the above step, an outlier in the time series is identified using an autocorrelation function, a moving average model; the abnormal values comprise diseases and insect pests, extreme weather or other external factors, and the identification of the abnormal values is helpful for timely taking intervention measures;
Then, short-term and long-term growth situation prediction is carried out by using an ARIMA model and an LSTM model;
the ARIMA model formula is:
wherein phi is i And theta i Is a model parameter, L is a hysteresis operator, d is the number of differences, ε t Is an error term;
finally, analyzing the correlation between the time series data of plant growth and the time series data of other environmental factors; the correlation adopts a Pearson correlation coefficient; for measuring a linear correlation between two consecutive variables;
wherein X and Y are respectively two time series data,and->Is their average value.
Further, the spectral data training AI model adopts the following steps:
s1, data set processing: normalizing the spectral data to ensure that the model has the same sensitivity at each frequency; the standardized processing formula is adopted:
wherein X is the original data, and mu and sigma are the mean value and standard deviation of the data respectively;
in the implementation process, in order to increase the generalization capability of the model, random small disturbance, noise addition or window sliding skills are used for enhancing the original spectrum data;
s2, model architecture:
input layer: receiving normalized one-dimensional spectrum data;
Convolution layer: capturing local patterns and features in the spectral data using a plurality of 1D convolution filters; employing an activation function including a ReLU to increase the nonlinearity of the model; the calculation includes the following:
f(x)=max(0,x)
pooling layer: using maximum pooling to reduce the dimensionality of the convolutional layer output while maintaining key features;
full tie layer: after the convolving and pooling layers, performing classification tasks using one or more fully connected layers; the number of the nodes of the last full connection layer is matched with the target classification number;
s3, loss function and optimization:
loss function: evaluating an error between the model output and the real label using a cross entropy loss function;
the calculation includes the following:
regularization: to prevent overfitting, the model adds L2 regularization during training:
where λ is the regularization parameter and w is the weight of the model;
optimization algorithm: an Adam optimizer is adopted, so that training is more stable and rapid;
m t =β 1 m t-1 +(1-β 1 )g t
wherein m is t And v t First and second moment estimates, g, respectively t Is the gradient at time t;
s4, early stop and model verification:
early-stop strategy: stopping training when the performance of the validation set does not significantly improve in consecutive iterations to avoid overfitting;
model verification: after each training period is finished, evaluating the performance of the model by using the verification set; and adjusting the model architecture, regularization strength or learning rate according to the performance feedback.
Further, the technical link of three-dimensional scanning of plants by using a camera or a laser radar is described in the following steps:
s1, three-dimensional data acquisition:
a. using a camera: capturing images of plants from different angles by using two cameras by using a binocular stereo camera, and estimating depth by using differential parallax; a structured light camera is adopted, a known light mode is projected onto plants, and depth information is calculated from the reflected mode;
b. laser radar was used: the lidar generates three-dimensional coordinates for each laser point by transmitting laser pulses and measuring the time of the pulses reflected back from the object to estimate the distance;
s2, three-dimensional data processing:
generating point cloud data: firstly, converting data acquired by a camera and a laser radar into point cloud data;
and (3) processing point cloud data: removing noise points, smoothing and space downsampling;
three-dimensional reconstruction: creating a continuous three-dimensional surface model from the point cloud data using techniques including a curved surface reconstruction algorithm;
s3, simulating plant growth:
parameterized model: constructing a plant growth model based on three-dimensional data; the plant growth model is a parameterized model, wherein parameters represent growth rate and branching mode;
And (3) simulated growth: simulating an expected growth process of the plant using the model; predicting plant growth under different conditions by changing model parameters;
s4, comparing growth data:
actual data acquisition: three-dimensional scanning of the plant at a plurality of time points of the actual growth cycle;
simulation and actual data comparison: comparing the simulated plant growth data with the actually collected data;
and (3) error calculation: calculating a difference between the simulated data and the actual data; including using Mean Square Error (MSE) to quantify the difference between the two:
wherein Y is predicted Is simulated growth data, and Y actual Is the growth data actually collected;
s5, growth situation assessment:
situation judgment: based on the comparison result of the simulation and the actual data, evaluating the growth situation of the plant; the method has the advantages that the fruit difference is small, which indicates that the growth state of the plant is good; including large differences in fruits, further analysis is required to determine growth problems;
advice and adjustment: based on the assessment of growth conditions, suggestions and schemes are provided for the maintenance and management of plants, including adjustment of irrigation, fertilization, or other growth conditions.
Further, when the AI model successfully analyzes the growth situation of the plant, the following detailed implementation steps are as follows:
S1, collecting and arranging planting data:
environmental data: including soil humidity, temperature, pH, and illumination intensity;
plant growth data: including height, number of leaves, leaf color;
management measure history data: including irrigation, fertilization, time and method of pest control;
s2, establishing an intelligent recommendation algorithm model:
characteristic engineering: selecting and extracting important characteristics related to the growth situation;
model selection: a decision tree, a random forest, a gradient lifting tree or a neural network model is selected, and the decision tree, the random forest, the gradient lifting tree or the neural network model is selected according to the complexity of the problem and the property of data;
model training: training a recommendation algorithm model using the collected data;
s3, generating planting suggestions:
environmental advice: based on environmental data, including advice to adjust soil moisture or to use a specific type of fertilizer to improve soil PH;
the management measures suggest: based on historical data and current growth conditions, including suggesting increased or decreased irrigation, or suggesting specific pest control measures;
predicting future growth: predicting a growth situation in a future period of time based on the current data and the prediction model, including predicting the number of blades;
s4, a user interaction interface:
Input interface: allowing the user to enter or upload current plant and environmental data;
output interface: displaying the growth situation analysis result of the AI model and planting suggestions of the intelligent recommendation system;
feedback system: allowing the user to feedback the validity of the suggestion thereby helping to improve the recommendation system.
Further, when an intelligent recommendation system for providing planting suggestions for gardeners or farmers is constructed, firstly, performing feature engineering, performing depth analysis on features including climate, soil and plant growth, and performing proper coding and expansion; then, introducing a complex model structure comprising collaborative filtering and deep learning models, and combining prediction results of a plurality of models by adopting a model integration technology;
to further optimize the model, performing an optimization of the superparameter, customizing a special loss function, and applying special optimization techniques including gradient clipping or dropout; to ensure continuous optimization and user satisfaction of the model, a real-time feedback system is established, and real-time updating of the model is performed by combining online learning and reinforcement learning; finally, user grouping and user portraits establishment are carried out to meet the specific requirements of different users so as to provide more accurate and personalized planting suggestions;
Further, the growth situation of the plant under different environmental conditions is predicted, firstly, a meteorological sensor and a soil sensor are deployed in a specific greening area or a field, the sensors monitor key indexes of air temperature, humidity, illumination, soil humidity, pH value, conductivity and soil temperature in real time, and all data are transmitted to a centralized data acquisition module for reading, preliminary processing and storage;
in the data preprocessing stage, cleaning abnormal data and carrying out standardization processing; further, carrying out feature engineering on the data, extracting key features including average daily temperature and soil humidity mean value, and dividing the data set into a training set, a verification set and a test set;
in the model establishment stage, selecting a model predicted by a time sequence to include LSTM or GRU for training, and performing model tuning by using a verification set; finally, inputting current and past environmental data into the model, predicting the future plant growth situation, and providing targeted agricultural advice according to real-time feedback;
further, the analysis of plant images using deep learning methods involves a number of steps: firstly, collecting plant images by adopting multi-angle equipment comprising a camera, and then carrying out data enhancement comprising rotation and color conversion;
Each image needs to be accurately marked to indicate the growth stage and disease type of the plant; then, a pre-training model comprising ResNet or VGG is used for extracting characteristics, performing migration learning and fine-tuning plant data;
adding batch normalization during model training, selecting a loss function, using an optimizer comprising Adam, and combining a learning rate attenuation strategy; when the accuracy of the model is evaluated, cross verification is used, the confusion matrix is analyzed, and fine adjustment is carried out aiming at the category with high misjudgment; finally, deploying the model to edge equipment for real-time monitoring, and setting a real-time feedback system, allowing a user to verify and optimize the model, and ensuring that the AI model can automatically identify and evaluate the growth, disease and pest states of plants;
further, the method for identifying abnormal values in the time series analysis of plant growth adopts the following steps: the autocorrelation function measures the correlation of the time series under different hysteresis values, and the abnormality is found out according to the captured periodic pattern;
then, calculating a local mode of the time series by using a moving average model, and determining an abnormal point based on residual calculation between the local mode and an actual value; further, the abnormal points are compared with external events, including weather records or pest outbreaks, so as to confirm the actual cause of the abnormality; finally, combining these identified anomalies, provides accurate intervention advice to farmers or gardeners, including the use of specific pesticides or adjustment of irrigation strategies, while predicting and warning of possible anomalies in the future, ensuring stable and healthy growth of plants.
The invention has the beneficial effects that:
1. high precision evaluation: by combining image recognition, spectral analysis and environmental sensor data, the method can provide a comprehensive and accurate assessment of plant growth situation. This is greatly superior to conventional visual inspection and empirical determination, and can effectively reduce false positives and false negatives.
2. Early intervention: the AI model may analyze the data in real time or near real time to identify potential growth problems, diseases or pest infestations early, thereby providing early intervention advice to gardeners or farmers to reduce plant losses.
3. Optimizing resource usage: by accurately identifying the needs of plants, farmers or gardeners can more accurately allocate resources, such as water, fertilizer and pesticides, thereby improving resource utilization efficiency and reducing waste.
4. And (3) an automatic flow: the method can realize automatic collection, processing and analysis of data, greatly reduce the amount of manual labor and improve the production efficiency.
5. Flexible adaptability: the AI model can continuously learn and self-adjust, is suitable for different environments and plant types, and has strong universality and adaptability.
6. Long-term trend analysis: the AI is used for analyzing the long-term data, so that gardeners and farmers can be helped to know the growth trend, the periodical change and the influence of external factors of plants, and data support is provided for future decisions.
7. Enhancing ecological sustainability: through accurate fertilization, irrigation and pest control, environmental pollution is reduced, and ecological sustainability is improved.
8. The cost is reduced: unnecessary pesticides and fertilizers are reduced, and losses caused by plant diseases and pests are avoided, so that the cost is saved for farmers and gardeners.
Drawings
FIG. 1 is a flow chart of a data acquisition and analysis method for identifying the growth situation of greening plants by using AI.
Detailed Description
The following describes the embodiments of the present invention in detail with reference to the drawings.
Examples: example 1: taking the growth monitoring of "yew" as an example, the following is a detailed explanation of the above steps:
s1: a multimodal sensor located in the yew park is used, wherein the RGB camera captures the color change of the plant, the infrared camera monitors its leaf temperature, the temperature and humidity sensor records the climate conditions, and the soil moisture sensor measures the soil moisture. For example, an infrared camera detects that leaf temperatures of certain plants are higher than surrounding plants, which means that these plants are affected by the disease. The neural network fuses the data to generate a growth health score for each Taxus chinensis.
S2: continuously recorded yew growth height data (e.g., 30cm, 31cm, 32cm height per day) were entered into a time series analysis model. The model recognizes a pattern in which taxus chinensis grows 2cm per month, and predicts the growth of the following month.
S3: the spectrometer measures spectral reflectance data of the taxus chinensis leaves. For example, healthy Taxus leaves have a specific spectral peak at 680nm, and leaves affected by pests will have this peak reduced. After training the AI model, these subtle spectral differences can be identified and the health of the plant can be assessed accordingly.
S4: and carrying out three-dimensional scanning on the taxus chinensis by using a camera to obtain a complete three-dimensional structure of the taxus chinensis. These data can help simulate future growth of yew during implementation. For example, the height, the number of branches, etc. of yew in the future three months are predicted in the implementation process at the current growth rate.
S5: the environmental sensor records various environmental conditions for the growth of taxus chinensis. For example, the temperature is 22 ℃, the humidity is 60%, the illumination intensity is 8000lux, and the soil humidity is 70%. Using these data, the AI model predicts that, for example, the growth rate of yew will slow down when humidity rises to 80%.
S6: the deep learning model represents the presence of leaf spot disease by learning images of a large number of yew, for example, spots on leaves, while insects of a certain shape and color are pests. After the model is trained, the characteristics can be automatically identified, and the gardener or farmer can be helped to take measures in time.
S7: the intelligent recommendation system analyzes all the data and provides specific planting suggestions for gardeners or farmers. For example, the system suggests: "leaf temperature is high, it is recommended to examine the yew of row 12 for the presence of disease. Irrigation was reduced to prevent further spread of the disease. "
Example 2: taking the growth monitoring of photinia fraseri as an example, the following detailed explanation is made:
s1, configuring sensors and collecting data: a group of sensors are arranged in the photinia fraseri garden, wherein an RGB camera shoots images of photinia fraseri, and a temperature and humidity sensor records the temperature and humidity of the growing environment. For example, on the afternoon of a certain day, the RGB camera captures an image of photinia fraseri, while the temperature and humidity sensor records a temperature of 28 ℃ and a humidity of 70%.
S2, data preprocessing: the image data is affected by ambient light, and overexposure or underexposure occurs; the temperature and humidity data are affected by sensor errors. Image data is subjected to brightness and contrast adjustment by using histogram equalization; the temperature and humidity data are filtered out of the spike noise through a sliding window method.
S3, feature extraction: image data: morphology features of photinia fraseri are extracted from RGB images using Convolutional Neural Networks (CNNs). For example, the trained CNN can identify the size, color and texture of photinia fraseri and output a feature vector F 1 =[0.8,0.3,0.7](representing size, color and texture).
Temperature and humidity data: environmental trends are extracted from temperature and humidity data over consecutive days using time series analysis, such as moving averages. For example, the average temperature over the past three days is 26 ℃,27 ℃,28 ℃, forming the feature vector F 2 =[26,27,28]。
S4, feature fusion: according to the above calculation expression, the method can be implemented by adding (F 1 ) And (F) 2 ) And (5) fusion. Given that the influence of the known image data on the growth situation of photinia fraseri in the implementation process is greater than the temperature and humidity, w is given 1 =0.7,w 2 Weight of =0.3. The fused features are expressed as:
F=0.7×[0.8,0.3,0.7]+0.3×[26,27,28]
F=[0.86,8.1,8.5]
s5, training an AI model: an AI model, such as a random forest, support vector machine, etc., is trained using the historical data and the fusion features to predict the growth situation of photinia fraseri.
S6, dynamically adjusting: if the AI model finds that the growth situation of photinia fraseri has larger deviation from the prediction on a certain day, the fusion strategy can be adjusted, such as increasing the weight of temperature and humidity data, because the growth environment of photinia fraseri is reflected more accurately.
Example 3: taking "nursery" growth monitoring as an example, the following detailed explanation is made:
s1, data collection: the nursery height was measured daily, assuming that data was collected continuously for one year during the course of the run.
Time: 365 days
Average height of nursery: 1150cm (throughout the year, from planting to harvesting)
S2, data cleaning:
data missing on day 50 and day 51 was found during the run, which can be filled using interpolation methods. Assuming a 30cm day 49 and a 34cm day 52, a 31cm day 50 and a 32cm day 51 may be assumed during the implementation.
S3, fourier transformation:
the data in the implementation process has obvious seasonal patterns, such as fast growth in summer. Using fourier transform, the implementation procedure yields:
wherein a is 0 =75 (average), other coefficient a n And b n Representing seasonal fluctuations in the data.
S4, linear regression and moving average:
assuming that the long-term trend was found to be linear in the course of the implementation, the height of the nursery was increased by about 0.4cm per day. Using moving averages, smoother trend lines may be obtained during implementation, reducing daily fluctuations.
S5, identifying abnormal values:
by means of the autocorrelation function and the moving average model, a sudden decrease of 10cm in height on day 100 was found in the implementation. This is due to external factors such as pests. This sudden drop requires further investigation.
S6.ARIMA model:
future growth was predicted using ARIMA, obtained during the course of the implementation:
(1φ 1 L)X t =μ+ε t
for example, if phi 1 =0.5, μ=0.4, which means that daily growth is based on half of the previous day and a constant growth rate.
S7, correlation analysis: it is assumed that daily temperature data Y are also collected during the implementation. In practice, it is desirable to know whether the temperature is related to nursery growth.
It is assumed that the number of the sub-blocks,and->(average temperature)
Using Pearson correlation coefficient formula, the implementation yields a value of r, such as r=0.8, which means that the temperature is highly correlated with the growth height of the nursery.
Example 4: an item is being carried out in order to identify different green plants by means of spectroscopic data. Spectral data of oak and pine were collected during the implementation.
S1, data set processing:
s1.1, standardization: assuming that the original value obtained from spectral data at a particular frequency during implementation is x=150, the mean μ=100 and the standard deviation σ=20, using the above normalization formula, we obtain:
s1.2, data enhancement: and randomly disturbing the original data. For example, if the original data is 150, a random noise in the range of 5 to 5 is added, and the implementation results in 152 or 147, etc.
S2, model architecture:
s2.1, input layer: one-dimensional spectral data of length 1000 is received.
S2.2. convolution layer: 10 1D convolution filters, each of length 5, are used to identify local patterns in the spectral data.
S2.3. activation function: using the ReLU function, for example, the outputs of ReLU are 0 and 3 for inputs 2 and 3.
S2.4, pooling layer: maximum pooling of length 2 is used, halving the dimension of the convolutional layer output.
S2.5, full connection layer: a full connection layer with 256 nodes is used first, and then a full connection layer with 2 nodes (oak and pine) is used.
S3, loss function and optimization:
s3.1. loss function:
assuming that the model predicts an output of [0.9,0.1] for spectral data of an oak, and the actual label is [1,0], using the cross entropy formula, the loss is:
L=-(1*log(0.9)+0*log(0.1))
s3.2, regularization: assuming λ=0.01, the sum of the model weights w is 50, and the regularization term is:
L regularization =0.01*50 2
s3.3.adam optimizer: in each training step, m is updated using a gradient t And v t To adjust the weight of the model.
S4, early stop and model verification:
s4.1, early-stop strategy: assuming 10 cycles of endurance is set during the exercise, if performance on the validation set is not improved during 10 consecutive training cycles, training is stopped during the exercise.
S4.2, model verification: assuming that 20% of the data is segmented during the implementation as a validation set, the performance of the model will be evaluated on this portion of the data during the implementation after each training period has ended.
Example 5: some agronomic research institute wants to study the growth status of sunflower and its response to different environmental factors by three-dimensional scanning techniques.
S1, three-dimensional data acquisition:
a. using a camera: the institute is equipped with binocular stereo camera and monitors the sunflower. When the sunflower grows to a height of 10cm, the two cameras capture their images from a position 15cm apart. The resulting disparities are used to calculate depth information due to their positional differences.
b. Laser radar was used: meanwhile, the laser radar emits laser pulses to the sunflower, and three-dimensional coordinates are generated according to the reflection time.
S2, three-dimensional data processing:
generating point cloud data: the data acquired from the cameras and lidar is converted into point cloud data, including millions of data points.
And (3) processing point cloud data: noise generated by the swing of the plant caused by wind is removed using software and smoothed.
Three-dimensional reconstruction: a curved reconstruction algorithm is applied, generating a continuous three-dimensional model from the point cloud data, which model shows the exact shape of the sunflower stems and leaves.
S3, simulating plant growth:
parameterized model: a sunflower growth model based on three-dimensional data was constructed, where the growth rate was 2 cm/day and the branching pattern produced one leaf per 5cm growth.
And (3) simulated growth: model predicts that within the next 10 days, sunflower will grow to 30cm and 4 new leaves will be produced.
S4, comparing growth data:
actual data acquisition: for the next 10 days, sunflowers were scanned three-dimensionally every day.
Simulation and actual data comparison: after 10 days, the actual data showed that the sunflower height was 28cm, 2cm from the model predicted 30 cm.
And (3) error calculation: using the MSE equation described above, an error value is calculated.
S5, growth situation assessment:
situation judgment: since the difference between the model and the actual growth data is small, the sunflower growth state can be considered to be good.
Advice and adjustment: however, to get closer to the actual growth rate, the institute decided to increase the soil humidity from 60% to 65% by adding a bit of irrigation on the next day.
Conclusion: through the steps, the farm institute can accurately monitor the growth situation of plants and provide more accurate planting suggestions for farmers, so that the yield and quality of flower buds are improved.
Example 6: some farmers want to increase the yield of their gardenia flower buds by using AI technology.
S1, collecting and arranging planting data:
environmental data:
soil humidity: 65%
Temperature: 25 DEG C
pH value: 6.8
Illumination intensity: 2000lux
Plant growth data:
height: 35cm
Number of blades: 20
Leaf color: dark green
Management measure history data:
irrigation: once a week, 100ml each time
And (3) fertilization: once a month, compound fertilizer is used
And (3) pest control: spraying insecticide once in last month
S2, establishing an intelligent recommendation algorithm model:
characteristic engineering:
according to historical data analysis, the soil humidity, the PH value and the illumination intensity have strong correlation with the growth of the gardenia jasminoides ellis.
Model selection:
due to the presence of multiple input features and possible interactions, a random forest model is selected for training.
Model training:
model training was performed using the collected data described above and historical gardenia jasminoides ellis growth data.
S3, generating planting suggestions:
environmental advice:
the soil humidity is increased to 70% to achieve the optimal growth condition of the gardenia jasminoides ellis.
The PH of the soil was slightly raised to 7.0 using an alkaline fertilizer.
The management measures suggest:
irrigation was increased to 150ml twice a week.
The use of biopesticides, rather than chemical pesticides, is contemplated to reduce the negative impact on the soil.
Predicting future growth:
if the above-mentioned recommended actions are followed, the number of leaves of Gardenia jasminoides Ellis is expected to increase to 30.
S4, a user interaction interface:
input interface:
the user can input current plant and environment data through a simple drop down menu and input box.
Output interface: the predicted growth situation of the gardenia jasminoides ellis is shown, including the predicted number of leaves and the planting proposal.
Feedback system: the user may click on the "suggest if valid" button, select "valid" or "invalid" and provide specific feedback.
Example 7: an intelligent planting assistant APP aims at providing personalized planting suggestions for farmers and gardeners.
1. Characteristic engineering:
climate characteristics: including average temperature, rainfall, humidity and wind speed. For example, the average temperature in a week in a certain area is 25 ℃, the rainfall is 20mm, the humidity is 65%, and the wind speed is 3m/s.
Soil characteristics: such as pH, water content, organic content, etc. For a particular soil sample, we may have a pH of 6.8, a water content of 40% and an organic content of 5%.
Plant growth characteristics: such as the number of days planted, plant height and leaf number. For a plant of one month of age, its height was 30cm and the number of leaves was 15.
Coding and expansion is performed, for example, for climate characteristics, seasonal characteristics, weather types (sunny, rainy), etc. can be generated.
2. Complex model structure:
collaborative filtering: to recommend planting recommendations to farmers that are similar to other farmers, for example, if farmers a and B have both chosen to plant a plant and have good results, they may be interested in the other recommendation.
Deep learning model: the above characteristics are processed by using a neural network, and the growth result of the plant is predicted. Complex relationships between features are captured through multiple hidden layers.
Model integration: techniques such as random forests, boosting, or Stacking are used in combination with the predictions of multiple models to improve accuracy.
3. Optimizing a model:
super parameter optimization: the best hyper-parameters are found using techniques such as mesh searching or bayesian optimization.
Special loss function: for example, using a weighted loss function, different weights are assigned to different planting suggestions, ensuring that the predictions of the key suggestions are more accurate.
Optimizing skills: for example, using dropout prevents overfitting, gradient clipping prevents gradient explosion.
4. Real-time feedback system:
when farmers make recommendations they can provide feedback, e.g. "these recommendations are very helpful" or "i tried these recommendations, but the effect was poor".
In combination with online learning and reinforcement learning, the model can be adjusted in real time based on these feedback to optimize subsequent advice.
5. User grouping and user portrayal:
farmers are divided into different groups, for example, a "fruit tree planter" or an "organic vegetable planter", by a clustering algorithm.
User portraits are created for each group, ensuring that the planting suggestions provided match their needs and experience.
This intelligent planting assistant APP can provide accurate and personalized planting advice for farmers and gardeners, helping them to plant more efficiently and increase harvesting.
Example 8: the paddy field in a certain area aims to predict the growth situation of the lawn and the targeted agricultural advice through real-time monitoring and data analysis.
1. Deploying a sensor:
meteorological and soil sensors are deployed in paddy fields. The data for one month are as follows (only parts are listed):
the data is transmitted to a centralized data acquisition module for processing and storage.
2. Data preprocessing:
Cleaning abnormal data: for example, if the illumination data for a day is significantly lower than for other days (e.g., 1 lux), this may be a sensor error, requiring removal or repair of this data.
Normalization processing: converting all data ranges to between 0 and 1 ensures that the magnitude of one feature does not affect the training of the model.
3. Characteristic engineering:
the average daily temperature and the average of the soil humidity were calculated.
The data sets are divided into training sets (e.g., 8/18/20), validation sets (8/2-18/25) and test sets (8/26-8/30).
4. And (3) establishing a model:
LSTM is used for training because it is particularly effective in processing time series data. Such data may be used as input features, considering that air temperature, humidity, etc. may affect the growth of the lawn.
Model parameters, such as the number of layers, the number of hidden units, etc., of the LSTM are adjusted using the verification set until the model performs optimally on the verification set.
5. Prediction and advice:
environmental data of 8/26 and 8/30 are input, and the growth situation of the lawn in the next days is predicted. For example, the model may predict that lawn growth will slow down during the next three days due to increased air temperature and decreased soil humidity.
Based on these predictions, the system may provide the following agricultural recommendations: taking into account the rise in air temperature and the fall in soil humidity, it is recommended to water twice daily for three days in the future and avoid fertilization during the high temperature period.
Example 9: in a nursery garden, it is desirable to analyze images of nursery plants by deep learning methods to identify and evaluate their growth, disease and pest status.
1. Image collection:
the multi-angle cameras are fixed at different positions of the nursery garden, and images of the nursery plants are automatically captured every day. For example, 5000 pictures may be collected within a month.
2. Data enhancement:
to increase the robustness of the model, each picture is rotated (e.g., by 90 °, 180 °, 270 °), color transformed (increasing/decreasing brightness, saturation change, etc.). Thus, 5000 pictures may be extended to 20000.
3. Image marking:
each image is accurately labeled using a special image labeling tool, such as: "primary leaf blade", "mature stage", "leaf spot" or "red spider attack".
4. Feature extraction and migration learning:
and adopting a pre-trained ResNet model to extract the characteristics of the nursery pictures. Since ResNet is trained on ImageNet, its weights can be used as initial values. Next, transfer learning is performed, training only the last layers of networks, and fine tuning is performed for nursery data.
5. Model training:
Batch normalization is used, which helps the network converge faster. Cross entropy was chosen as the loss function and Adam optimizer was used. In order to prevent the learning rate from stagnating, a learning rate decay strategy is adopted, so that the learning rate is gradually reduced in the training process.
6. Evaluating model accuracy:
cross-validation was used to evaluate the performance of the model. For example, if the model has problems identifying "red spider attacks", this can be clarified by confusion matrices.
7. Deployment and real-time monitoring:
the trained model is deployed to an edge device in the garden, such as a camera-equipped Raspberry Pi. The gardener is automatically notified when the camera detects a characteristic similar to a disease or pest.
8. Real-time feedback system:
if the campus caller finds that the determination of the AI model is incorrect, it can be fed back to the system via a simple interface. This feedback can be used to further fine tune and optimize the model, ensuring its performance in a real environment.
Conclusion: through the steps, the gardener can automatically identify and evaluate the growth, diseases and pest states of the nursery by relying on the AI model, so that the workload of manual inspection is greatly reduced, and the early detection and treatment efficiency of the pests and diseases is improved.
Example 10: a study person observed a red maple field and recorded the growth height of red maple every day. After several months of data collection we have obtained a time dependent height data sequence.
1. Autocorrelation function:
the autocorrelation function may reveal internal structures of time series data, such as seasonal and trending. For example, if a significant autocorrelation peak is observed every 7 days, this may indicate a weekly periodicity.
From the autocorrelation map we may observe that the autocorrelation at some point is exceptionally high, except for the expected periodic peak. These points may be outliers.
2. Moving average model:
this model can help us better understand the local pattern of the time series. For example, if we calculate a 7-day moving average, we may notice that some of the actual data points deviate significantly from their corresponding moving averages, which may also be abnormal.
3. In contrast to external events:
after finding the outlier, it is compared with the local weather record or pest report. For example, the growth of a certain red maple suddenly slows down, while the weather record of the day shows an abnormal rise in temperature. Thus, we can conclude that high temperature may be responsible for affecting red maple growth.
4. Providing an intervention suggestion:
in combination with the above analysis, intervention advice is provided to farmers. For example, if temperature is found to be the cause of the effect on growth, farmers may be advised to provide more moisture or shade on high temperature days or advise them to use specific pesticides against possible pests.
At the same time, the system may predict anomalies that may occur in the future, such as predicting that there may be a slower growth rate of one day in the future week based on past data.
Example 11: the object is: a comprehensive plant growth monitoring and health management system is constructed, so that real-time acquisition, analysis and prediction of plant growth data are realized, and scientific and accurate agricultural advice is provided for farmers or gardeners.
1. Data acquisition and pretreatment:
and (3) equipment deployment: and (3) deploying a meteorological sensor and a soil sensor in the field, and monitoring key indexes such as air temperature, humidity, illumination, soil humidity, pH value, conductivity, soil temperature and the like. Plant images are collected from multiple angles using a camera.
Data transmission and storage: all data are transmitted to a central data acquisition module in real time for storage and preliminary processing.
Pretreatment: data cleaning, outlier processing and normalization. The image data is enhanced, including rotation, color conversion, and the like.
2. Time series analysis and outlier identification:
the periodic pattern of the time series is measured by an autocorrelation function and possible anomalies are found from the periodic pattern.
The local pattern of the time series is calculated using a moving average model, and anomalies are identified based on differences in actual values from the moving average.
And comparing the abnormal data with external weather records or pest outbreak records to determine the actual cause of the abnormality.
3. Deep learning image analysis:
feature extraction of plant images is performed using pre-trained models, such as ResNet or VGG. And performing migration learning and fine tuning on the plant image data.
Batch normalization, specific loss functions and optimizers are added in the model training process, and learning rate attenuation strategies are combined.
And the plant growth stage and the disease type are automatically identified through the deep learning model.
4. Prediction and suggestion:
based on the environmental data, a time series prediction model (e.g., LSTM or GRU) is used to predict future plant growth conditions.
In combination with the above analysis, agricultural advice is provided to farmers or gardeners, such as adjusting irrigation strategies, selecting appropriate pesticides, etc.
The early warning system can predict the possible abnormal conditions of the plants according to analysis and inform farmers in advance.
5. Real-time feedback and system optimization:
and deploying a deep learning model on the edge equipment for real-time monitoring.
A real-time feedback system is built and the user can verify the predicted results of the model and provide feedback to help further optimize the model.
By combining on-line learning and reinforcement learning, the system can update the model in real time according to the received feedback, and accuracy is improved.
Summarizing: the embodiment provides a comprehensive and real-time plant growth monitoring and management tool for farmers or gardeners by integrating various technical means, helps them to more scientifically and accurately culture and manage plants, and ensures healthy growth of the plants.
The foregoing has shown and described the basic principles, principal features and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, and that the above embodiments and descriptions are merely illustrative of the principles of the present invention, and various changes and modifications may be made without departing from the spirit and scope of the invention, which is defined in the appended claims. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (10)
1. The data acquisition and analysis method for identifying the growth situation of the greening plants by using the AI is characterized by comprising the following steps of:
s1, collecting plant growth data from multiple angles by utilizing a multi-mode sensor comprising an RGB camera, an infrared camera, a temperature and humidity sensor and a soil humidity sensor, and fusing the sensor data by a neural network method;
s2, performing time sequence analysis on the obtained continuous plant growth data, identifying a growth curve and a periodic mode, and predicting a future growth situation according to the past data;
s3, collecting spectral reflection data of plant leaves by utilizing a spectrometer, and training an AI model based on the spectral data to identify and evaluate different growth states and health conditions of plants;
s4, three-dimensional scanning is carried out on the plants by using a camera or a laser radar, three-dimensional data of the plants are obtained, the growth process of the plants is further simulated by using the three-dimensional data, and the growth situation of the plants is estimated by comparing the growth process with actual growth data;
s5, collecting environmental factor data related to plant growth, including air temperature, humidity, illumination and soil conditions, by integrating an environmental sensor, and predicting growth situations of plants under different environmental conditions by combining the environmental factor data and using an AI model;
S6, applying a deep learning method, and enabling the AI model to automatically identify and evaluate the growth stage, diseases and pest invasion states of plants through training a large number of plant images;
s7, further constructing an intelligent recommendation system on the basis of analyzing plant growth situation by the AI model, and providing targeted planting suggestions for gardeners or farmers.
2. The data acquisition and analysis method for identifying the growth situation of the greening plants by using the AI according to claim 1, wherein the multi-mode sensor data processing process is as follows: configuring a set of sensors to systematically collect data of plants and their environments from a plurality of angles and dimensions; preprocessing data from different sources, eliminating noise, and performing standardization and normalization operation; extracting key features related to plant growth situation from data of each mode by using a neural network structure, wherein the key features comprise morphological features of plants extracted from RGB images by using a convolutional neural network or environmental change features extracted from temperature and humidity data by using a time sequence analysis method;
according to the demands of plant growth situations, adopting weighting, stacking or cascading strategies to integrate the characteristics of different modes into a unified characteristic representation; training an AI model by using the fused features, and dynamically adjusting the strategy and weight of data fusion according to the data collected in real time and the evaluation result of the AI model;
The weighting strategy is calculated by the following expression:
F=w 1 ×F 1 +w 2 ×F 2
wherein F is a fused feature, F 1 And F 2 Is characteristic of two modes, w 1 And w 2 Is their weight.
3. The method for collecting and analyzing data for identifying the growth situation of a greening plant according to claim 1, wherein the time series analysis is to continuously collect time series data related to the growth of the plant, including factors of the height, the number of leaves, the area of the leaves and the environment in which the plants are located; then, data cleaning is carried out, wherein the data cleaning comprises filling of missing values and abnormal value removal;
then using fourier transforms or seasonal decomposition to identify periodic patterns in the time series; the specific formula is as follows:
wherein X (t) is a time series, a 0 ,a n ,b n Is a coefficient, N is a period, f 0 Is the fundamental frequency;
then, linear regression and moving average technology are used for identifying long-term trend, so that the growth situation of the plant in the future is predicted;
in the above step, an outlier in the time series is identified using an autocorrelation function, a moving average model; the abnormal values comprise diseases and insect pests, extreme weather or other external factors, and the identification of the abnormal values is helpful for timely taking intervention measures;
Then, carrying out short-term and long-term growth situation prediction by using an ARIMA model;
the ARIMA model formula is:
wherein phi is i And theta i Is a model parameter, L is a hysteresis operator, d is the number of differences, ε t Is an error term;
finally, analyzing the correlation between the time series data of plant growth and the time series data of other environmental factors; the correlation adopts a Pearson correlation coefficient; for measuring a linear correlation between two consecutive variables;
wherein X and Y are respectively two time series data,and->Is their average value.
4. The data acquisition and analysis method for identifying the growth situation of the greening plants by using the AI according to claim 1, wherein the spectral data training AI model comprises the following steps:
s1, data set processing: normalizing the spectral data to ensure that the model has the same sensitivity at each frequency; the standardized processing formula is adopted:
wherein X is the original data, and mu and sigma are the mean value and standard deviation of the data respectively;
in the implementation process, in order to increase the generalization capability of the model, random small disturbance, noise addition or window sliding skills are used for enhancing the original spectrum data;
S2, model architecture:
input layer: receiving normalized one-dimensional spectrum data;
convolution layer: capturing local patterns and features in the spectral data using a plurality of 1D convolution filters; employing an activation function including a ReLU to increase the nonlinearity of the model; the calculation includes the following:
f(x)=max(0,x)
pooling layer: using maximum pooling to reduce the dimensionality of the convolutional layer output while maintaining key features;
full tie layer: after the convolving and pooling layers, performing classification tasks using one or more fully connected layers; the number of the nodes of the last full connection layer is matched with the target classification number;
s3, loss function and optimization:
loss function: evaluating an error between the model output and the real label using a cross entropy loss function;
the calculation includes the following:
regularization: to prevent overfitting, the model adds L2 regularization during training:
where λ is the regularization parameter and w is the weight of the model;
optimization algorithm: an Adam optimizer is adopted, so that training is more stable and rapid;
m t =β 1 m t-1 +(1-β 1 )g t
wherein m is t And v t First and second moment estimates, g, respectively t Is the gradient at time t;
s4, early stop and model verification:
early-stop strategy: stopping training when the performance of the validation set does not significantly improve in consecutive iterations to avoid overfitting; model verification: after each training period is finished, evaluating the performance of the model by using the verification set; and adjusting the model architecture, regularization strength or learning rate according to the performance feedback.
5. The data acquisition and analysis method for identifying the growth situation of the greening plants by using the AI according to claim 1, wherein the data acquisition and analysis method comprises the following steps: the technical link of three-dimensional scanning of plants by using cameras or laser radars is described in the following steps:
s1, three-dimensional data acquisition:
a. using a camera: capturing images of plants from different angles by using two cameras by using a binocular stereo camera, and estimating depth by using differential parallax; a structured light camera is adopted, a known light mode is projected onto plants, and depth information is calculated from the reflected mode;
b. laser radar was used: the lidar generates three-dimensional coordinates for each laser point by transmitting laser pulses and measuring the time of the pulses reflected back from the object to estimate the distance;
s2, three-dimensional data processing:
generating point cloud data: firstly, converting data acquired by a camera and a laser radar into point cloud data;
and (3) processing point cloud data: removing noise points, smoothing and space downsampling;
three-dimensional reconstruction: creating a continuous three-dimensional surface model from the point cloud data using techniques including a curved surface reconstruction algorithm;
s3, simulating plant growth:
parameterized model: constructing a plant growth model based on three-dimensional data; the plant growth model is a parameterized model, wherein parameters represent growth rate and branching mode;
And (3) simulated growth: simulating an expected growth process of the plant using the model; predicting plant growth under different conditions by changing model parameters;
s4, comparing growth data:
actual data acquisition: three-dimensional scanning of the plant at a plurality of time points of the actual growth cycle;
simulation and actual data comparison: comparing the simulated plant growth data with the actually collected data;
and (3) error calculation: calculating a difference between the simulated data and the actual data; including using Mean Square Error (MSE) to quantify the difference between the two:
wherein Y is predicted Is simulated growth data, and Y actual Is the growth data actually collected;
s5, growth situation assessment:
situation judgment: based on the comparison result of the simulation and the actual data, evaluating the growth situation of the plant; the method has the advantages that the fruit difference is small, which indicates that the growth state of the plant is good; including large differences in fruits, further analysis is required to determine growth problems;
advice and adjustment: based on the assessment of growth conditions, suggestions and schemes are provided for the maintenance and management of plants, including adjustment of irrigation, fertilization, or other growth conditions.
6. The data acquisition and analysis method for identifying the growth situation of a green plant by using an AI according to claim 1, wherein the following steps are implemented in detail when the AI model successfully analyzes the growth situation of the plant:
S1, collecting and arranging planting data:
environmental data: including soil humidity, temperature, pH, and illumination intensity;
plant growth data: including height, number of leaves, leaf color;
management measure history data: including irrigation, fertilization, time and method of pest control;
s2, establishing an intelligent recommendation algorithm model:
characteristic engineering: selecting and extracting important characteristics related to the growth situation;
model selection: a decision tree, a random forest, a gradient lifting tree or a neural network model is selected, and the decision tree, the random forest, the gradient lifting tree or the neural network model is selected according to the complexity of the problem and the property of data;
model training: training a recommendation algorithm model using the collected data;
s3, generating planting suggestions:
environmental advice: based on environmental data, including advice to adjust soil moisture or to use a specific type of fertilizer to improve soil PH; the management measures suggest: based on historical data and current growth conditions, including suggesting increased or decreased irrigation, or suggesting specific pest control measures;
predicting future growth: predicting a growth situation in a future period of time based on the current data and the prediction model, including predicting the number of blades;
s4, a user interaction interface:
input interface: allowing the user to enter or upload current plant and environmental data;
Output interface: displaying the growth situation analysis result of the AI model and planting suggestions of the intelligent recommendation system;
feedback system: allowing the user to feedback the validity of the suggestion thereby helping to improve the recommendation system.
7. The data acquisition and analysis method for identifying the growth situation of greening plants by using the AI according to claim 6, wherein when an intelligent recommendation system for providing planting suggestions for gardeners or farmers is constructed, characteristic engineering is firstly performed, characteristics including climate, soil and plant growth are deeply analyzed, and proper coding and expansion are performed; then, introducing a complex model structure comprising collaborative filtering and deep learning models, and combining prediction results of a plurality of models by adopting a model integration technology;
to further optimize the model, performing an optimization of the superparameter, customizing a special loss function, and applying special optimization techniques including gradient clipping or dropout; to ensure continuous optimization and user satisfaction of the model, a real-time feedback system is established, and real-time updating of the model is performed by combining online learning and reinforcement learning; finally, user grouping and user portrait establishment are carried out to meet the specific requirements of different users so as to provide more accurate and personalized planting suggestions.
8. The data acquisition and analysis method for identifying the growth situation of the greening plants by using the AI according to claim 1 is characterized in that the growth situation of the greening plants under different environmental conditions is predicted, a meteorological sensor and a soil sensor are deployed in a specific greening area or field, the sensors monitor key indexes of air temperature, humidity, illumination, soil humidity, pH value, conductivity and soil temperature in real time, and all data are transmitted to a centralized data acquisition module for reading, preliminary processing and storage;
in the data preprocessing stage, cleaning abnormal data and carrying out standardization processing; further, carrying out feature engineering on the data, extracting key features including average daily temperature and soil humidity mean value, and dividing the data set into a training set, a verification set and a test set;
in the model establishment stage, selecting a model predicted by a time sequence to include LSTM or GRU for training, and performing model tuning by using a verification set; finally, current and past environmental data are input into the model, future plant growth situations are predicted, and targeted agricultural advice is provided according to real-time feedback.
9. The data acquisition and analysis method for identifying the growth situation of the greening plants by using the AI according to claim 1, wherein the analysis of the plant images by using the deep learning method involves a plurality of steps: firstly, collecting plant images by adopting multi-angle equipment comprising a camera, and then carrying out data enhancement comprising rotation and color conversion;
Each image needs to be accurately marked to indicate the growth stage and disease type of the plant; then, a pre-training model comprising ResNet or VGG is used for extracting characteristics, performing migration learning and fine-tuning plant data;
adding batch normalization during model training, selecting a loss function, using an optimizer comprising Adam, and combining a learning rate attenuation strategy; when the accuracy of the model is evaluated, cross verification is used, the confusion matrix is analyzed, and fine adjustment is carried out aiming at the category with high misjudgment; and finally, deploying the model to edge equipment for real-time monitoring, and setting a real-time feedback system, so that a user is allowed to verify and optimize the model, and the AI model is ensured to automatically identify and evaluate the growth, disease and pest states of plants.
10. The data acquisition and analysis method for identifying the growth situation of the greening plants by using the AI according to claim 1, wherein the abnormal value identification method in the time sequence analysis of the plant growth is characterized by comprising the following steps: the autocorrelation function measures the correlation of the time series under different hysteresis values, and the abnormality is found out according to the captured periodic pattern;
then, calculating a local mode of the time series by using a moving average model, and determining an abnormal point based on residual calculation between the local mode and an actual value; further, the abnormal points are compared with external events, including weather records or pest outbreaks, so as to confirm the actual cause of the abnormality; finally, combining these identified anomalies, provides accurate intervention advice to farmers or gardeners, including the use of specific pesticides or adjustment of irrigation strategies, while predicting and warning of possible anomalies in the future, ensuring stable and healthy growth of plants.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311051190.5A CN117036088A (en) | 2023-08-21 | 2023-08-21 | Data acquisition and analysis method for identifying growth situation of greening plants by AI |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311051190.5A CN117036088A (en) | 2023-08-21 | 2023-08-21 | Data acquisition and analysis method for identifying growth situation of greening plants by AI |
Publications (1)
Publication Number | Publication Date |
---|---|
CN117036088A true CN117036088A (en) | 2023-11-10 |
Family
ID=88644691
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311051190.5A Pending CN117036088A (en) | 2023-08-21 | 2023-08-21 | Data acquisition and analysis method for identifying growth situation of greening plants by AI |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117036088A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117251004A (en) * | 2023-11-17 | 2023-12-19 | 金杲易光电科技(深圳)有限公司 | PH value control system for plant growth |
CN117370823A (en) * | 2023-12-05 | 2024-01-09 | 恒健达(辽宁)医学科技有限公司 | Spraying control method and system for agricultural planting |
CN117516639A (en) * | 2024-01-08 | 2024-02-06 | 吉林农业大学 | High-flux greenhouse plant phenotype measurement system based on multispectral point cloud fusion |
CN117743975A (en) * | 2024-02-21 | 2024-03-22 | 君研生物科技(山西)有限公司 | Hillside cultivated land soil environment improvement method |
-
2023
- 2023-08-21 CN CN202311051190.5A patent/CN117036088A/en active Pending
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117251004A (en) * | 2023-11-17 | 2023-12-19 | 金杲易光电科技(深圳)有限公司 | PH value control system for plant growth |
CN117370823A (en) * | 2023-12-05 | 2024-01-09 | 恒健达(辽宁)医学科技有限公司 | Spraying control method and system for agricultural planting |
CN117370823B (en) * | 2023-12-05 | 2024-02-20 | 恒健达(辽宁)医学科技有限公司 | Spraying control method and system for agricultural planting |
CN117516639A (en) * | 2024-01-08 | 2024-02-06 | 吉林农业大学 | High-flux greenhouse plant phenotype measurement system based on multispectral point cloud fusion |
CN117743975A (en) * | 2024-02-21 | 2024-03-22 | 君研生物科技(山西)有限公司 | Hillside cultivated land soil environment improvement method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US10349584B2 (en) | System and method for plant monitoring | |
Khairunniza-Bejo et al. | Application of artificial neural network in predicting crop yield: A review | |
CN117036088A (en) | Data acquisition and analysis method for identifying growth situation of greening plants by AI | |
US20170161560A1 (en) | System and method for harvest yield prediction | |
US20220075344A1 (en) | A method of finding a target environment suitable for growth of a plant variety | |
WO2022253057A1 (en) | Intelligent moisture precise irrigation control system and method for fruit and vegetable cultivation in solar greenhouse | |
CN111008733B (en) | Crop growth control method and system | |
CN112906298B (en) | Blueberry yield prediction method based on machine learning | |
CN111476149A (en) | Plant cultivation control method and system | |
Lin et al. | Intelligent greenhouse system based on remote sensing images and machine learning promotes the efficiency of agricultural economic growth | |
CN110119169A (en) | A kind of tomato greenhouse temperature intelligent early warning system based on minimum vector machine | |
Singh et al. | A systematic review of artificial intelligence in agriculture | |
Veeragandham et al. | A review on the role of machine learning in agriculture | |
CN116151454A (en) | Method and system for predicting yield of short-forest linalool essential oil by multispectral unmanned aerial vehicle | |
Lin et al. | Data-driven modeling for crop growth in plant factories | |
Sharma et al. | Crop yield prediction using hybrid deep learning algorithm for smart agriculture | |
Araneta et al. | Controlled Environment for Spinach Cultured Plant with Health Analysis using Machine Learning | |
CN115438934A (en) | Crop growth environment monitoring method and system based on block chain | |
US20230054908A1 (en) | Machine learning optimization through randomized autonomous crop planting | |
AlKameli et al. | Automatic learning in agriculture: a survey | |
Chaudhary et al. | A critical review on hybrid framework for precise farming with application of Machine Learning (ML) and Internet of Things (IoT) | |
FAISAL | A pest monitoring system for agriculture using deep learning | |
CN111640039A (en) | Facility vegetable disease control recommendation system and method | |
Jasti et al. | Crop Intelligent: Weather based Crop Selection using Machine Learning | |
Priya et al. | A survey on deep learning based IoT approach for precision crop suggestion |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |