Mountain tobacco field production situation prediction method based on image recognition

Abstract

The invention discloses a mountain tobacco field production situation prediction method based on image recognition. The method comprises the following steps: step 1, collecting image information through a camera, collecting meteorological breath and soil information through various sensors, and regularly uploading the collected data to a data cloud through a remote communication module by a control module; 2, directly splicing or aligning the collected original data, and inputting the data into a data analysis model; and step 3, constructing a data analysis model from three aspects of yield prediction, quality grading and disaster risk early warning, and generating training data which are diversified and conform to mountain agricultural laws by adopting two data enhancement strategies of mountain environment simulation and sample generation. Image and soil information is creatively added, various information in the tobacco leaf growth process is monitored in real time in an all-around mode, a data analysis model and a data display platform are developed in a matched mode, and possibility is provided for digital transformation of the tobacco leaf growth link.

Description

Technical Field

[0001] The present invention relates to the technical field of Internet of Things in tobacco production, and particularly to a method for predicting the production situation of mountain tobacco fields based on image recognition. Background Technique

[0002] Tobacco is a kind of solanaceous plant, native to the Americas, and its leaves contain alkaloids such as nicotine. In addition to being made into cigarettes, sun-cured tobacco, pipe tobacco, cigars, etc. for people to smoke, tobacco also has a variety of medical uses. Although tobacco has brought many harms to humans, as a medicinal plant with a long history, its medical value cannot be denied due to its harmfulness.

[0003] The role of tobacco field monitoring is to track the tobacco growth environment, pest and disease dynamics, and production management effects in real time through technical means, aiming to improve yield, quality and sustainability, especially in mountain tobacco-growing areas with fragile ecosystems. Traditional tobacco field monitoring mainly relies on tobacco technicians to check in the fields. In some places, meteorological monitoring stations have been developed, but their structures are relatively simple and the monitoring data is relatively one-sided. Therefore, we propose a method for predicting the production situation of mountain tobacco fields based on image recognition. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the existing defects, provide a method for predicting the production situation of mountain tobacco fields based on image recognition, creatively add image and soil information, monitor various information in the tobacco leaf growth process in real time and all-round, and develop a data analysis model and a data display platform in a supporting manner, providing the possibility for the digital transformation of the tobacco leaf growth link, and effectively solving the problems in the background technique.

[0005] To achieve the above object, the present invention provides the following technical solution: A method for predicting the production situation of mountain tobacco fields based on image recognition, including the following steps;

[0006] Step 1: Collect image information through a camera, collect meteorological and soil information through various sensors, and the control module regularly uploads the collected data to the data cloud through the remote communication module;

[0007] Step 2: Directly splice or align the collected original data and input it into the data analysis model;

[0008] Step 3: Construct a data analysis model from three aspects of yield prediction, quality grading and disaster risk warning, and adopt two data augmentation strategies of mountain environment simulation and sample generation to generate training data that is both diverse and in line with mountain agronomic laws, and finally make the production situation prediction model adapt to the complexity of the real world;

[0009] Step 4: Generate growth curves and heatmaps through the data analysis model, create key charts and provide farming suggestions through the data display platform, creatively add image and soil information, comprehensively monitor various information in real time during the tobacco leaf growth process, and develop a data analysis model and a data display platform in a supporting manner, providing the possibility for the digital transformation of the tobacco leaf growth link.

[0010] Furthermore, yield prediction, quality grading, and disaster risk warning are the core functional modules of the data analysis model. The output unit of yield prediction is kg / mu, which is generated based on the vegetation indices (NDVI, EVI) and plant density in the multispectral images. Quality grading divides grades according to the maturity of tobacco leaves and the proportion of damaged patches (such as the 42-grade system of national standards). The disaster risk warning synchronously outputs the probability of pests and diseases (such as the infestation rate of Myzus persicae) and the drought index. The three functional modules complement each other in their functions and jointly serve the scientific management of tobacco cultivation and the optimization of economic benefits.

[0011] Furthermore, the data display platform integrates sensor data, meteorological information, historical yield records, etc., eliminates "data islands", forms a digital twin of the tobacco field, and based on the model prediction results, the platform automatically pushes action suggestions, such as disaster warnings: marking high-risk plots (such as red areas: infestation rate of Myzus persicae > 40%), recommending biological control solutions, and pushing the best operation time according to the growth stage (seedling raising, transplanting, harvesting). The data display platform is the key bridge connecting technological achievements and practical applications, realizing visual data, and more importantly, empowering the whole-chain decision-making of tobacco cultivation through multi-dimensional interaction and intelligent analysis.

[0012] Furthermore, the image information collected by the camera is three types of image information: visible light images, multispectral images, and thermal infrared images. Visible light images are used to collect the growth status of tobacco leaves, providing information such as leaf color, density, and plant height, providing a basis for subsequent image analysis and processing. Multispectral images feedback chlorophyll content and water stress index, and thermal infrared images monitor canopy temperature and warn of drought.

[0013] Furthermore, the various sensors collect meteorological information through meteorological information such as temperature and humidity sensors, wind speed, wind direction, and rainfall, and can generate temperature and humidity curves and calculate the accumulated temperature duration. High temperature and high humidity are likely to induce various diseases, and temperature and humidity data can give timely warnings. Through soil information such as soil temperature and humidity and soil nitrogen, phosphorus, and potassium content, the soil moisture content and the content of nutrients such as nitrogen, phosphorus, and potassium during the tobacco leaf growth process can be grasped in real time.

[0014] Furthermore, the sample generation adopts methods of geometric transformation, color adjustment, and noise injection. Geometric transformation involves rotating the sample (±15°), flipping, and cropping (retaining the main body of the tobacco plant). Color adjustment is to adjust the brightness (±30%), contrast (±20%), and saturation (±15%) of the sample. Noise injection is to add Gaussian noise (σ = 0.05) and salt-and-pepper noise (density 1%) to enhance data diversity and alleviate overfitting. Sample generation is a key technical means to solve data scarcity, class imbalance, and generalization in complex environments.

[0015] Furthermore, the mountain environment simulation uses an atmospheric scattering model (such as the Koschmi eder model) to generate fog effects according to visibility parameters (1 - 5 km), uses Particle Systems to generate rain filaments with random directions and densities, superimposes them on the image, and simulates tobacco field images with different slopes through rotation, tilting, and scaling, which is applicable to complex mountain terrains.

[0016] Furthermore, the sample generation adopts methods of geometric transformation, color adjustment, and noise injection. Geometric transformation involves rotating the sample (±15°), flipping, and cropping (retaining the main body of the tobacco plant). Color adjustment is to adjust the brightness (±30%), contrast (±20%), and saturation (±15%) of the sample. Noise injection is to add Gaussian noise (σ = 0.05) and salt-and-pepper noise (density 1%) to make the yield prediction model adapt to the complexity of the real world.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: The method for predicting the yield of mountain tobacco fields based on image recognition has the following advantages:

[0018] 1. Automatically collect meteorological information, soil information, and image data during the growth process of tobacco leaves through various sensors, which has the advantages of reducing the labor intensity of tobacco technicians and monitoring the growth elements of tobacco leaves in real time throughout the process, providing support for the digital management of tobacco fields.

[0019] 2. Construct a data analysis model based on the collected data. The data analysis model is optimized through two data augmentation strategies: mountain environment simulation and sample generation, and displays information such as the collected data and key charts through a data display platform, which can guide tobacco farmers to reasonably arrange farming activities such as watering and fertilizing, and at the same time can provide disease warnings, take measures in advance, and reduce tobacco leaf losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic diagram of the architecture of the present invention;

[0021] Figure 2 It is a schematic diagram of the composition of the data cloud of the present invention;

[0022] Figure 3Schematic diagram of the composition of the data acquisition module of the present invention;

[0023] Figure 4 Schematic diagram of the operation process of the present invention. Specific implementation manners

[0024] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0025] Please refer to Figures 1-4 , this embodiment provides a technical solution: a method for predicting the yield situation of mountain tobacco fields based on image recognition, including the following steps;

[0026] Step 1: Collect image information through a camera, and various sensors collect meteorological and soil information. The control module regularly uploads the collected data to the data cloud through the remote communication module. The image information collected by the camera is three types of image information: visible light image, multispectral image, and thermal infrared image. The visible light image is used to collect the growth status of tobacco leaves, providing information such as leaf color, density, plant height, etc., providing a basis for subsequent image analysis and processing. The multispectral image feedbacks the chlorophyll content and water stress index, and the thermal infrared image monitors the canopy temperature to warn of drought. The various sensors collect meteorological information such as temperature and humidity sensors, wind speed, wind direction, and rainfall, etc., which can generate temperature and humidity curves and calculate the accumulated temperature duration. High temperature and high humidity are likely to induce various diseases, and the temperature and humidity data can give early warnings in time. Through soil information such as soil temperature and humidity and soil nitrogen, phosphorus, and potassium content, the soil moisture content and the content of nutrients such as nitrogen, phosphorus, and potassium during the growth process of tobacco leaves can be grasped in real time. The soil temperature and humidity detection adopts dynamic adjustment, and the buried depth of the soil temperature and humidity meter is adjusted according to the growth period of tobacco. It is shallowly buried at 10-15 cm during the transplanting period, mainly buried at 20-25 cm during the vigorous growth period, and deeply buried at 30 cm during the mature period;

[0027] Step 2: Directly splice or align the collected original data and then input it into the data analysis model;

[0028] Step 3: Construct a data analysis model from three aspects: yield prediction, quality grading, and disaster risk warning. And adopt two data augmentation strategies: mountain environment simulation (randomly adding cloud, fog, rain filaments, slope affine transformation) and sample generation (using geometric transformation, color adjustment, and noise injection methods) to generate diverse training data that conforms to mountain agronomic laws. Eventually, make the production situation prediction model adapt to the complexity of the real world. Yield prediction, quality grading, and disaster risk warning are the core functional modules of the data analysis model. The output unit of yield prediction is kg / mu, which is generated based on the vegetation indices (NDVI, EVI) and plant density in the multispectral image. Quality grading divides grades according to the maturity of tobacco leaves and the proportion of damaged patches (such as the 42-grade system of national standards). Disaster risk warning synchronously outputs the probability of pests and diseases (such as the infestation rate of Myzus persicae) and drought index. The mountain environment simulation uses an atmospheric scattering model (such as the Koschmieder model) to generate fog effects according to the visibility parameter (1 - 5 km), uses Particle Systems to generate rain filaments with random directions and densities, superimposes them on the image, and simulates tobacco field images with different slopes through rotation, tilting, and scaling, which is applicable to the complex terrain of mountains. Sample generation adopts geometric transformation, color adjustment, and noise injection methods. Geometric transformation rotates the sample (±15°), flips it, and crops it (retaining the main body of the tobacco plant). Color adjustment adjusts the brightness (±30%), contrast (±20%), and saturation (±15%) of the sample. Noise injection adds Gaussian noise (σ = 0.05) and salt-and-pepper noise (density 1%);

[0029] Step 4: Generate growth curves and heatmaps through the data analysis model, and produce key charts and provide farming suggestions through the data display platform. The data display platform integrates sensor data, meteorological information, historical yield records, etc., eliminates "data islands", forms a digital twin of the tobacco field, and based on the model prediction results, the platform automatically pushes action suggestions, such as disaster warning: marking high-risk plots (such as the red area: infestation rate of Myzus persicae > 40%), recommending biological control solutions, and pushing the best operation time according to the growth stage (seedling raising, transplanting, harvesting).

[0030] The working principle of a method for predicting the production situation of mountain tobacco fields based on image recognition provided by the present invention is as follows: First, plan a suitable location in the mountain tobacco field, place earthworms and pour the base with concrete. Install a vertical rod on the upper surface of the base. A cross arm is provided at the upper end of the vertical rod, and a temperature and humidity sensor, a wind speed sensor, a wind direction sensor, a rain gauge, and a camera are installed at the upper end of the cross arm. The soil temperature and humidity meter and the soil nitrogen, phosphorus, and potassium meter are buried in the tobacco field. The soil temperature and humidity meter adjusts the burial depth according to the tobacco growth cycle. The data of various sensors are transmitted to the single-chip microcomputer and regularly sent to the data cloud through the remote module for data collection. After organizing the collected original data, it is uploaded to the data analysis model. Two data augmentation strategies, namely mountain environment simulation (randomly adding cloud and fog noise, slope affine transformation) and sample generation (using geometric transformation, color adjustment, and noise injection methods), are used to generate training data that is both diverse and in line with mountain agricultural laws. The data analysis model generates a growth curve and a heat map based on the input original data. Relevant personnel make key charts and provide agricultural suggestions, and display them through the data display platform.

[0031] The above are only the embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied to other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. A method for predicting tobacco field yield in mountainous areas based on image recognition, characterized in that: The steps include: Step 1: The camera collects image information, various sensors collect meteorological information and soil information, and the control module regularly uploads the collected data to the data cloud through the remote communication module; Step 2: directly splice or align the collected raw data and input them into the data analysis model; Step 3: Build a data analysis model from three aspects: yield prediction, quality grading, and disaster risk warning. Use two data enhancement strategies, mountain environment simulation and sample generation, to generate training data that is both diverse and in line with the laws of mountain agronomy, ultimately making the yield prediction model adapt to the complexity of the real world. Step 4: Generate growth curves and heat maps through data analysis models, and create key charts and provide farming advice through the data display platform.

2. The method for predicting the yield of mountain tobacco fields based on image recognition according to claim 1, characterized in that: The yield prediction, quality grading and disaster risk warning are the core functional modules of the data analysis model. The output unit of yield prediction is kg / mu, which is generated according to the vegetation index (NDVI, EVI) and plant density in the multispectral image. The quality grading is divided into grades according to the maturity of tobacco leaves and the proportion of damaged patches (such as the 42-grade system of the national standard). The disaster risk warning simultaneously outputs the probability of pests and diseases (such as the infection rate of tobacco aphids) and the drought index.

3. The method for predicting tobacco field yield in mountainous areas based on image recognition according to claim 1, characterized in that: The data display platform eliminates "data islands" by integrating sensor data, meteorological information, historical production records, etc., forming a digital twin of the tobacco field. Based on the model prediction results, the platform automatically pushes action suggestions, such as disaster warnings: marking high-risk plots (such as red zones: tobacco aphid infestation rate >40%), recommending biological control plans, and pushing the best operation time according to the growth stage (seedling raising, transplanting, harvesting).

4. The method for predicting tobacco field yield in mountainous areas based on image recognition according to claim 1, characterized in that: The camera collects three types of image information: visible light images, multispectral images and thermal infrared images. The visible light images are used to collect the growth status of tobacco leaves, provide information such as leaf color, density, plant height, etc., and provide a basis for subsequent image analysis and processing. The multispectral images feedback chlorophyll content and water stress index, and the thermal infrared images monitor canopy temperature and warn of drought.

5. The method for predicting tobacco field yield in mountainous areas based on image recognition according to claim 1, characterized in that: The various sensors collect meteorological information through thermometers, humidity meters, wind speed, wind direction, rainfall and other meteorological information, which can generate temperature and humidity curves and calculate the accumulated temperature time. High temperature and high humidity can easily induce various diseases. Temperature and humidity data can provide timely warnings. Through soil information such as soil temperature and humidity and soil nitrogen, phosphorus and potassium content, the soil moisture conditions and the content of nutrients such as nitrogen, phosphorus and potassium during tobacco leaf growth can be grasped in real time.

6. The method for predicting tobacco field yield in mountainous areas based on image recognition according to claim 5, characterized in that: The soil temperature and humidity detection adopts dynamic adjustment, and the burying depth of the soil temperature and humidity meter is adjusted according to the tobacco growth period, with a shallow burial of 10-15cm during the transplanting period, a main burial of 20-25cm during the vigorous period, and a deep burial of 30cm during the maturity period.

7. The method for predicting tobacco field yield in mountainous areas based on image recognition according to claim 1, characterized in that: The mountain environment simulation uses an atmospheric scattering model (such as the Koschmieder model) to generate fog effects according to visibility parameters (1-5km), uses Particle Systems to generate raindrops of random direction and density, superimposes them on the image, and simulates tobacco field images of different slopes through rotation, tilting, and scaling.

8. The method for predicting tobacco field yield in mountainous areas based on image recognition according to claim 1, characterized in that: The sample generation adopts the methods of geometric transformation, color adjustment and noise injection. The geometric transformation is to rotate (±15°), flip and crop the sample (retain the main body of the tobacco plant). The color adjustment is to adjust the brightness (±30%), contrast (±20%) and saturation (±15%) of the sample. The noise injection is to add Gaussian noise (σ=0.05) and salt and pepper noise (density 1%).

Images