WO2023024615A1 - 一种基于植株萎蔫程度进行精准灌溉的系统及方法 - Google Patents
一种基于植株萎蔫程度进行精准灌溉的系统及方法 Download PDFInfo
- Publication number
- WO2023024615A1 WO2023024615A1 PCT/CN2022/094812 CN2022094812W WO2023024615A1 WO 2023024615 A1 WO2023024615 A1 WO 2023024615A1 CN 2022094812 W CN2022094812 W CN 2022094812W WO 2023024615 A1 WO2023024615 A1 WO 2023024615A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- image
- information
- plant
- neural network
- water demand
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000003973 irrigation Methods 0.000 title claims abstract description 29
- 230000002262 irrigation Effects 0.000 title claims abstract description 29
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 78
- 238000011176 pooling Methods 0.000 claims abstract description 57
- 238000013528 artificial neural network Methods 0.000 claims abstract description 30
- 238000007781 pre-processing Methods 0.000 claims abstract description 30
- 238000004364 calculation method Methods 0.000 claims abstract description 9
- 241000196324 Embryophyta Species 0.000 claims description 70
- 238000013527 convolutional neural network Methods 0.000 claims description 39
- 230000011218 segmentation Effects 0.000 claims description 11
- 238000013507 mapping Methods 0.000 claims description 8
- 230000002708 enhancing effect Effects 0.000 claims description 6
- 238000012216 screening Methods 0.000 claims description 6
- 238000012549 training Methods 0.000 abstract description 9
- 238000012545 processing Methods 0.000 abstract description 3
- 238000006243 chemical reaction Methods 0.000 abstract description 2
- 230000005284 excitation Effects 0.000 abstract 1
- 239000008400 supply water Substances 0.000 abstract 1
- 230000006870 function Effects 0.000 description 39
- 230000004913 activation Effects 0.000 description 12
- 238000003860 storage Methods 0.000 description 11
- 238000004422 calculation algorithm Methods 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 5
- 230000003287 optical effect Effects 0.000 description 4
- 230000001629 suppression Effects 0.000 description 4
- 235000013311 vegetables Nutrition 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 230000008034 disappearance Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 239000002689 soil Substances 0.000 description 3
- 230000015572 biosynthetic process Effects 0.000 description 2
- 210000004556 brain Anatomy 0.000 description 2
- 238000000354 decomposition reaction Methods 0.000 description 2
- 238000003709 image segmentation Methods 0.000 description 2
- 230000003902 lesion Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000003786 synthesis reaction Methods 0.000 description 2
- 235000002566 Capsicum Nutrition 0.000 description 1
- 240000008067 Cucumis sativus Species 0.000 description 1
- 235000009849 Cucumis sativus Nutrition 0.000 description 1
- 235000007688 Lycopersicon esculentum Nutrition 0.000 description 1
- 241000758706 Piperaceae Species 0.000 description 1
- 240000003768 Solanum lycopersicum Species 0.000 description 1
- 244000061458 Solanum melongena Species 0.000 description 1
- 235000002597 Solanum melongena Nutrition 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 239000003337 fertilizer Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000003621 irrigation water Substances 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008450 motivation Effects 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000001556 precipitation Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012552 review Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A01—AGRICULTURE; FORESTRY; ANIMAL HUSBANDRY; HUNTING; TRAPPING; FISHING
- A01G—HORTICULTURE; CULTIVATION OF VEGETABLES, FLOWERS, RICE, FRUIT, VINES, HOPS OR SEAWEED; FORESTRY; WATERING
- A01G25/00—Watering gardens, fields, sports grounds or the like
- A01G25/16—Control of watering
- A01G25/162—Sequential operation
Definitions
- the invention belongs to the technical field of agricultural plant irrigation, and in particular relates to a system and method for precise irrigation based on the wilting degree of plants.
- This application requires the application number submitted to the Patent Office of the State Intellectual Property Office of China on August 23, 2021: CN202110971230. 2 application, the entire disclosure of which is incorporated herein by reference in its entirety.
- Chinese patent document CN 109845625 A discloses a neural network-based multi-dimensional parameter intelligent irrigation control method for crops, by collecting parameters such as rainfall information, soil moisture information, wind speed information, temperature and humidity information, light intensity information and flow information of the current irrigated farmland. Based on the neural network, the crop water demand model is established based on the crop water demand signal as the response information. Through this model, the multi-dimensional environmental parameters of the crops are calculated and processed, and finally the water demand of the current farmland crops is predicted. The controller controls the water demand and rainfall. and soil moisture to make a comprehensive judgment result and control the solenoid valve according to the judgment result to realize the irrigation of crops.
- this technical scheme considers the factors affecting irrigation water demand more comprehensively, the model constructed is very simple, and the accuracy of water demand prediction results is not high.
- the present invention proposes to use the proposed deep neural network to carry out intelligent water precision irrigation control, which greatly improves the prediction accuracy of water demand, and simultaneously considers the water shortage characteristics of multiple leaves and stems to form Multiple belief networks are fused and judged, and the water demand of plants is obtained comprehensively.
- This application uses a system and method for precise irrigation based on the degree of wilting of plants to realize precise irrigation control of plants. According to the characteristics of water shortage in multiple leaves and stems, multiple confidence judgment networks are formed, and the comprehensive results of plant Water demand, to prevent the existence of special lesions on individual leaves from affecting the accuracy of judgment.
- this application adopts new algorithms in the use of preprocessing, segmentation, pooling layer, activation function and loss function, so as to improve the training of deep neural network as a whole accuracy and speed.
- preprocessing segmentation, pooling layer, activation function and loss function
- the G component is enhanced, and the R and B components are relatively suppressed, which is conducive to improving the accuracy of demand forecasting and judgment .
- the concept of deep learning was proposed by Hinton et al. in 2006. Based on the deep belief network (DBN), a non-supervised greedy layer-by-layer training algorithm is proposed, which brings hope to solve the optimization problems related to the deep structure, and then a multi-layer autoencoder deep structure is proposed.
- the convolutional neural network proposed by Lecun et al. is the first real multi-layer structure learning algorithm, which uses the spatial relative relationship to reduce the number of parameters to improve training performance.
- Deep learning is a new field in machine learning research. Its motivation is to establish and simulate the neural network of human brain for analysis and learning. It imitates the mechanism of human brain to explain data, such as images, sounds and texts. Like machine learning methods, deep machine learning methods can also be divided into supervised learning and unsupervised learning. The learning models established under different learning frameworks are very different. For example, Convolutional neural networks (CNNs) is a machine learning model under deep supervised learning, and Deep Belief Nets (DBNs) is a machine learning model under unsupervised learning. .
- CNNs Convolutional neural networks
- DNNs Deep Belief Nets
- CNN Convolutional Neural Networks
- Feedforward Neural Networks Feedforward Neural Networks
- It is one of the representative algorithms for deep learning.
- the deep convolutional neural network DCNN is a network structure with multiple CNN layers.
- the activation functions often used in deep neural networks are as follows: sigmoid function, tanh function, ReLU function.
- the sigmoid function which maps the value of (- ⁇ ,+ ⁇ ) to (0,1).
- the formula of the sigmoid function is as follows:
- the sigmoid function is used as a nonlinear activation function, but it is not often used. It has the following disadvantages:
- the tanh function is more common than the sigmoid function. This function maps the value of (- ⁇ ,+ ⁇ ) to (-1,1).
- the formula is:
- the tanh function can be regarded as linear in a short region around 0. Since the mean value of the tanh function is 0, it makes up for the shortcoming of the mean value of the sigmoid function being 0.5.
- the ReLU function also known as the Rectified Linear Unit, is a piecewise linear function that makes up for the gradient disappearance problem of the sigmoid function and the tanh function.
- the formula of the ReLU function is as follows:
- the present invention is a system for precise irrigation of facility vegetables based on the degree of wilting of plants.
- the system includes:
- the image acquisition module collects the image information of the plant through the image sensor, and the collected information includes: leaf information and stem information;
- An image preprocessing module the image preprocessing module is used to complete image preprocessing, including at least one of image cropping, segmentation, and enhancement;
- the water demand prediction module inputs the collected image information into the trained deep neural network for calculation, and predicts the corresponding predicted water demand under the current state of wilting degree of plants;
- the irrigation control module forms control instructions according to the predicted water demand of the plants, so as to provide precise and quantitative water supply.
- the preprocessing also includes: screening of image information, eliminating images that do not meet the definition requirements or fail to contain at least one complete leaf; the image clipping includes: clipping the obtained image information to obtain Leaf information and stem information, and establish the corresponding relationship between the leaf and the plant stem.
- the preprocessing further includes: performing color suppression on the R and B components in the RGB space, and enhancing the G channel component.
- the deep neural network specifically includes a multi-region convolutional neural network model
- the multi-regional convolutional neural network model includes: a deep convolutional neural network, used to generate mapping features of original leaves; multi-region confidence
- the network model including the confidence network model of multiple regions, is used to generate a plurality of different confidence values of different water demands for the current state of the plant, and fit the different confidence values of multiple regions to determine the If the confidence values in the area are all relatively large, the corresponding water demand is determined as the water demand of the plant.
- the multi-region belief network model includes a multi-region pooling layer and a fully connected layer
- the multi-region pooling layer includes pooling layers in multiple regions, the number of pooling layers is 1, and the pooling layer is also Can be replaced by a fully connected layer; the pooling layer is used to generate confidence.
- the present invention also proposes a method for precision irrigation based on the wilting degree of the plant, the method comprising:
- the image information of the plant is collected through the image sensor, and the collected information includes: leaf information and stem information;
- image preprocessing module to complete image preprocessing, including at least one of image cropping, segmentation, and enhancement;
- a control instruction is formed according to the predicted water demand of the plant, so as to provide precise and quantitative water supply.
- the preprocessing also includes: screening of image information, eliminating images that do not meet the definition requirements or fail to contain at least one complete leaf; the image clipping includes: clipping the obtained image information to obtain Leaf information and stem information, and establish the corresponding relationship between the leaf and the plant stem.
- the preprocessing further includes: performing color suppression on the R and B components in the RGB space, and enhancing the G channel component.
- the deep neural network specifically includes a multi-region convolutional neural network model
- the multi-regional convolutional neural network model includes: a deep convolutional neural network, used to generate mapping features of original leaves; multi-region confidence
- the network model including the confidence network model of multiple regions, is used to generate a plurality of different confidence values of different water demands for the current state of the plant, and fit the different confidence values of multiple regions to determine the If the confidence values in the area are all relatively large, the corresponding water demand is determined as the water demand of the plant.
- the multi-region belief network model includes a multi-region pooling layer and a fully connected layer
- the multi-region pooling layer includes pooling layers in multiple regions, the number of pooling layers is 1, and the pooling layer is also Can be replaced by a fully connected layer; the pooling layer is used to generate confidence.
- the present application also correspondingly proposes a computer storage medium, on which a program code is stored, and the code is used to implement any one of the methods described above.
- the present application also correspondingly proposes a computer device, the device includes a processor and a memory, and the memory stores computer instructions, and the instructions are used to implement any one of the methods described above.
- This application uses a system and method for precise irrigation based on the degree of wilting of plants to achieve precise irrigation control of plants. According to the characteristics of water shortage in multiple leaves and stems, multiple confidence judgment networks are formed to comprehensively obtain the plant Water demand, to prevent the existence of special lesions on individual leaves from affecting the accuracy of judgment.
- this application adopts new algorithms in the use of pooling layer, activation function and loss function, so as to improve the training accuracy and speed of deep neural network as a whole.
- this application adopts new algorithms in the use of pooling layer, activation function and loss function, so as to improve the training accuracy and speed of deep neural network as a whole.
- irrigation control it is the applicant's first proposal, so it is not a conventional technical means or common knowledge.
- the G component is enhanced, and the R and B components are relatively suppressed, which is conducive to improving the accuracy of demand forecasting and judgment .
- FIG. 1 shows a schematic structural view of a basic embodiment of the present application.
- the present application proposes a system for precision irrigation based on the degree of wilting of plants, the system includes:
- the image acquisition module collects the image information of the plant through the image sensor, and the collected information includes: leaf information and stem information;
- An image preprocessing module the image preprocessing module is used to complete image preprocessing, including at least one of image cropping, segmentation, and enhancement;
- the water demand prediction module inputs the collected image information into the trained deep neural network for calculation, and predicts the corresponding predicted water demand under the current state of wilting degree of plants;
- the irrigation control module forms a control command according to the predicted water demand of the plant, so as to provide precise and quantitative water supply.
- the preprocessing also includes: screening of image information, eliminating images that do not meet the definition requirements or fail to contain at least one complete leaf; the image clipping includes: clipping the obtained image information to obtain Leaf information and stem information, and establish the corresponding relationship between the leaf and the plant stem.
- the preprocessing further includes: performing color suppression on the R and B components in the RGB space, and enhancing the G channel component.
- the deep neural network specifically includes a multi-region convolutional neural network model
- the multi-regional convolutional neural network model includes: a deep convolutional neural network, used to generate mapping features of original leaves; multi-region confidence
- the network model including the confidence network model of multiple regions, is used to generate a plurality of different confidence values of different water demands for the current state of the plant, and fit the different confidence values of multiple regions to determine the If the confidence values in the area are all relatively large, the corresponding water demand is determined as the water demand of the plant.
- the multi-region belief network model includes a multi-region pooling layer and a fully connected layer
- the multi-region pooling layer includes pooling layers in multiple regions, the number of pooling layers is 1, and the pooling layer is also Can be replaced by a fully connected layer; the pooling layer is used to generate confidence.
- the G component is enhanced, and the R, G, and B components of the segmented leaves are first separated; the adjustment coefficient of the G component is generated:
- the decomposition and synthesis of each component belongs to the existing technology in this field, but the above adjustment method is the inventor's first creation. After the above adjustment, the green channel information of the blade can be fully utilized to make the subsequent prediction process more accurate.
- the image segmentation adopts an improved watershed segmentation method for segmentation:
- Gradient(x,y) represents the original gradient value of the pixel point (x,y); Respectively represent the average gradient value, minimum gradient value, and maximum gradient value in the window D area; Gra represents the corrected gradient value;
- the deep neural network includes a deep convolutional neural network and a multi-regional belief network, and the deep convolutional neural network is used to generate the segmented features of each leaf and stem;
- the multi-regional belief network includes The belief network model of multiple leaf regions and stem regions is used to generate confidence values for the degree of water shortage of multiple different leaves and stems, when the confidence values of at least two leaf regions and the confidence degrees of the corresponding stem regions When the values all meet the preset first threshold level range (corresponding to leaves) and second threshold level range (corresponding to stems), then the water demand of the plant is determined.
- the deep neural network includes a deep convolutional neural network and a multi-region belief network, and the deep convolutional neural network can receive the segmented leaf and stem images of the input plant, and generate convolutional neural networks of different scales.
- Multi-feature mapping ; the multi-region belief network model includes a multi-region pooling layer and a fully connected layer; wherein, the multi-region pooling layer includes a plurality of region pooling layers; the plurality of region pooling layers are used to generate The confidence value of the water demand of the plant image, when the confidence value of at least two leaf regions and the confidence value of the corresponding stem region all meet the preset first threshold level range (corresponding to the blade), the second threshold level range ( When corresponding to the stem), then determine the water demand of the plant; the multi-region pooling layer is set to the maximum pooling layer or the average pooling layer; the fully connected layer is used for water shortage of identified leaves and stems degree of classification.
- the deep neural network is a deep convolutional neural network, specifically including: an input layer, an embedding layer, a pooling layer, and a fully connected layer; the input layer is used to receive the segmented leaves and stems of the input plants Internal image;
- the convolution kernel size that described embedding layer adopts is 5*5;
- the activation function of the present invention is denoted as R l (); Further obtain water demand prediction recommendation result after fully connected layer processing;
- the pooling method of the pooling layer is as follows:
- x e represents the output of the current layer
- ue represents the input of the activation function R l
- R l () represents the activation function
- w e represents the weight of the current layer
- ⁇ represents the loss function
- x e-1 represents the output of the previous layer
- the activation function R l is:
- the loss function ⁇ is as follows:
- N represents the size of the positive sample data set, i takes a value from 1 to N, y i represents the label value corresponding to the positive sample x i ; W yi represents the weight of the positive sample feature vector x i at its label y i , and s is the depth volume
- the recommended parameters of the product neural network; b j represents the deviation of the sample xi at its label y i .
- the state of the stem can be measured by a stem flow sensor, instead of an image sensor to identify the state of the stem, and the measurement results of the runoff sensor can be input into a deep convolutional neural network for training and prediction.
- the present invention also proposes a method for precision irrigation based on the wilting degree of the plant, the method comprising:
- the image information of the plant is collected through the image sensor, and the collected information includes: leaf information and stem information;
- image preprocessing module to complete image preprocessing, including at least one of image cropping, segmentation, and enhancement;
- a control instruction is formed according to the predicted water demand of the plant, so as to provide precise and quantitative water supply.
- the preprocessing also includes: screening of image information, eliminating images that do not meet the definition requirements or fail to contain at least one complete leaf; the image clipping includes: clipping the obtained image information to obtain Leaf information and stem information, and establish the corresponding relationship between the leaf and the plant stem.
- the preprocessing further includes: performing color suppression on the R and B components in the RGB space, and enhancing the G channel component.
- the deep neural network specifically includes a multi-region convolutional neural network model
- the multi-regional convolutional neural network model includes: a deep convolutional neural network, used to generate mapping features of original leaves; multi-region confidence
- the network model including the confidence network model of multiple regions, is used to generate a plurality of different confidence values of different water demands for the current state of the plant, and fit the different confidence values of multiple regions to determine the If the confidence values in the area are all relatively large, the corresponding water demand is determined as the water demand of the plant.
- the multi-region belief network model includes a multi-region pooling layer and a fully connected layer
- the multi-region pooling layer includes pooling layers in multiple regions, the number of pooling layers is 1, and the pooling layer is also Can be replaced by a fully connected layer; the pooling layer is used to generate confidence.
- the G component is enhanced, and the R, G, and B components of the segmented leaves are first separated; the adjustment coefficient of the G component is generated:
- the decomposition and synthesis of each component belongs to the existing technology in this field, but the above adjustment method is the inventor's first creation. After the above adjustment, the green channel information of the blade can be fully utilized to make the subsequent prediction process more accurate.
- the image segmentation adopts an improved watershed segmentation method for segmentation:
- Gradient(x,y) represents the original gradient value of the pixel point (x,y); Respectively represent the average gradient value, minimum gradient value, and maximum gradient value in the window D area; Gra represents the corrected gradient value;
- the deep neural network includes a deep convolutional neural network and a multi-regional belief network, and the deep convolutional neural network is used to generate the segmented features of each leaf and stem;
- the multi-regional belief network includes The belief network model of multiple leaf regions and stem regions is used to generate confidence values for the degree of water shortage of multiple different leaves and stems, when the confidence values of at least two leaf regions and the confidence degrees of the corresponding stem regions When the values all meet the preset first threshold level range (corresponding to leaves) and second threshold level range (corresponding to stems), then the water demand of the plant is determined.
- the deep neural network includes a deep convolutional neural network and a multi-region belief network, and the deep convolutional neural network can receive the segmented leaf and stem images of the input plant, and generate convolutional neural networks of different scales.
- Multi-feature mapping ; the multi-region belief network model includes a multi-region pooling layer and a fully connected layer; wherein, the multi-region pooling layer includes a plurality of region pooling layers; the plurality of region pooling layers are used to generate The confidence value of the water demand of the plant image, when the confidence value of at least two leaf regions and the confidence value of the corresponding stem region all meet the preset first threshold level range (corresponding to the blade), the second threshold level range ( When corresponding to the stem), then determine the water demand of the plant; the multi-region pooling layer is set to the maximum pooling layer or the average pooling layer; the fully connected layer is used for water shortage of identified leaves and stems degree of classification.
- the deep neural network is a deep convolutional neural network, specifically including: an input layer, an embedding layer, a pooling layer, and a fully connected layer; the input layer is used to receive the segmented leaves and stems of the input plants Internal image;
- the convolution kernel size that described embedding layer adopts is 5*5;
- the activation function of the present invention is denoted as R l (); Further obtain water demand prediction recommendation result after fully connected layer processing;
- the pooling method of the pooling layer is as follows:
- x e represents the output of the current layer
- ue represents the input of the activation function R l
- R l () represents the activation function
- w e represents the weight of the current layer
- ⁇ represents the loss function
- x e-1 represents the output of the previous layer
- the activation function R l is:
- the loss function ⁇ is as follows:
- N represents the size of the positive sample data set, i takes a value from 1 to N, y i represents the label value corresponding to the positive sample x i ; W yi represents the weight of the positive sample feature vector x i at its label y i , and s is the depth volume
- the recommended parameters of the product neural network; b j represents the deviation of the sample xi at its label y i .
- the state of the stem can be measured by a stem flow sensor, instead of an image sensor to identify the state of the stem, and the measurement results of the runoff sensor can be input into a deep convolutional neural network for training and prediction.
- the present application also correspondingly proposes a computer storage medium, on which a program code is stored, and the code is used to implement any one of the methods described above.
- the present application also correspondingly proposes a computer device, the device includes a processor and a memory, and the memory stores computer instructions, and the instructions are used to implement any one of the methods described above.
- the present application also proposes a computer-readable medium, which contains program codes that can realize the above-mentioned system, and the contained program codes can be transmitted by any appropriate medium, including but not limited to wireless, electric wires, optical cables, RF, etc., or the above-mentioned any suitable combination.
- the computer readable medium may be a computer readable signal medium or a computer readable storage medium.
- a computer-readable storage medium may be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any combination thereof.
- a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.
- Computer program code for carrying out the operations of the present invention may be written in one or more programming languages, or combinations thereof, including object-oriented programming languages—such as Java, Smalltalk, C++, and conventional Procedural Programming Language - such as "C" or a similar programming language.
- the program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server.
- the remote computer can be connected to the user computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (such as via the Internet using an Internet service provider). connect).
- LAN local area network
- WAN wide area network
- connect such as via the Internet using an Internet service provider
- the above-mentioned integrated units implemented in the form of software functional units may be stored in a computer-readable storage medium.
- the above-mentioned software functional units are stored in a storage medium, and include several instructions to make a computer device (which may be a personal computer, server, or network device, etc.) or a processor (processor) execute the methods described in various embodiments of the present invention. partial steps.
- the aforementioned storage media include: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disc and other media that can store program codes. .
Landscapes
- Engineering & Computer Science (AREA)
- Water Supply & Treatment (AREA)
- Life Sciences & Earth Sciences (AREA)
- Environmental Sciences (AREA)
- Image Analysis (AREA)
Abstract
本申请提出一种基于植株萎蔫程度进行精准灌溉的系统及方法，包括：图像采集模块，通过图像传感器进行植株的图像信息采集，所采集的信息包括：叶片信息及茎部信息；图像预处理模块，所述图像预处理模块用于完成图像的预处理，包括图像裁剪、增强、色彩空间转化处理；需水量预测模块，将所采集的图像信息输入至经训练的深度神经网络进行计算，预测出当前植株萎蔫程度状态下对应的预测需水量；灌溉控制模块，根据预测出的所述植株的预测需水量形成控制指令，以进行精准定量的供水。本发明仅需识别叶片以及茎部特征便能准确估算出缺水量，而且利用所提出的损失函数、池化方法以及激励函数，能提升模型训练的速度以及预测的精度。
Description
本发明属于农业植物灌溉技术领域，尤其涉及一种基于植株萎蔫程度进行精准灌溉的系统及方法,本申请要求在2021年8月23日向中国国家知识产权局专利局提交的申请号为：CN202110971230.2的申请为优先权，其整个公开通过引用的方式全部并入此处。
我国园艺设施面积已达2840万亩，其中，日光温室约占31％，种植蔬菜种类主要包含辣椒、番茄、黄瓜、茄子等。设施反季节栽培已成为人们日常蔬菜供应重要的组成部分。但是，在实际生产过程中，生产者多凭借经验进行粗放灌溉，造成水资源浪费，降低肥料资源利用效率和果实品质。因此，实现基于蔬菜水分需求规律和外界环境精准灌溉对于节水提质变得尤为重要。
中国专利文献CN 109845625 A公开了一种基于神经网络的多维参量农作物智能灌溉控制方法，通过采集当前灌溉农田的雨量信息、土壤墒情信息、风速信息、温湿度信息、光照强度信息及流量信息等参量，基于神经网络建立以农作物需水信号为响应信息的农作物需水量模型，通过该模型对农作物多维环境参量进行计算处理，最终预测出当前农田农作物的需水量，控制器通过对需水量、降雨量及土壤墒情做出综合判决结果并根据判决结果控制电磁阀，实现对农作物的灌溉。该技术方案虽然考虑影响灌溉需水量的因素较全面，所构建的模型十分简单，需水量预测结果精度不高。
实际上，无论是降水量和/或供水量、土壤湿度等都会影响植株的水分需求量，但是植株缺水时，植株叶片以及茎部反应能较容易观察到或测定，因此将缺水量多少问题转化为对叶片以及茎部特征的识别问题，即可完成缺水量的预测问题。为解决上述问题，本发明提出利用所提出的深度神经网络进行智能水分精准灌溉控制，极大提高了需水量的预测精度，同时考虑多片叶子的缺水特征以及茎部的缺水特征，形成多个置信网络，对其进行融合判断，综合得出植株的需水量。
本申请的创造性贡献在于：
1.本申请利用一种基于植株萎蔫程度进行精准灌溉的系统及方法，实现了植株的精准灌溉控制，针对多个叶片和茎部缺水特征，形成多个置信度判断网络， 综合得出植株需水量，防止个别叶片存在特殊病变等情况影响判断的准确性。
2.本申请为了提升需水量的预测精度和训练速度，在预处理、分割、池化层、激励函数、损失函数的使用上，都采用了新的算法，以整体上提高深度神经网络的训练的精度和速度。在灌溉控制领域，属于申请人首次提出，因此并非常规技术手段或公知常识。
3.在叶片的预处理上，针对绝大部分植株叶片都是绿色的特点，对于识别到的叶片，增强其G分量，相对抑制其R、B分量，有利于提升需求量预测判断的准确性。
发明内容
为更准确理解本发明，需先简要理解回顾下面的基本概念。
深度学习的概念由Hinton等人于2006年提出。基于深信度网(DBN)提出非监督贪心逐层训练算法，为解决深层结构相关的优化难题带来希望，随后提出多层自动编码器深层结构。此外Lecun等人提出的卷积神经网络是第一个真正多层结构学习算法，它利用空间相对关系减少参数数目以提高训练性能。
深度学习是机器学习研究中的一个新的领域，其动机在于建立、模拟人脑进行分析学习的神经网络，它模仿人脑的机制来解释数据，例如图像、声音和文本。同机器学习方法一样，深度机器学习方法也有监督学习与无监督学习之分。不同的学习框架下建立的学习模型很是不同。例如，卷积神经网络(Convolutional neural networks，简称CNNs)就是一种深度的监督学习下的机器学习模型，而深度置信网(Deep Belief Nets，简称DBNs)就是一种无监督学习下的机器学习模型。
卷积神经网络(Convolutional Neural Networks,CNN)是一类包含卷积计算且具有深度结构的前馈神经网络(Feedforward Neural Networks)，是深度学习(deep learning)的代表算法之一。
深度卷积神经网络DCNN，则是具有多个CNN层的网络结构。
深度神经网络中经常采用的激励函数如下：sigmoid函数，tanh函数，ReLU函数。
sigmoid函数，该函数是将取值为(-∞,+∞)的数映射到(0,1)之间。sigmoid函数的公式如下：
sigmoid函数作为非线性激活函数，但是其并不被经常使用，它具有以下几个缺点：
(1)当z值非常大或者非常小时，sigmoid函数的导数g′(z)将接近0。这会导致权重W的梯度将接近0，使得梯度更新十分缓慢，即梯度消失。
tanh函数，tanh函数相较于sigmoid函数要常见一些，该函数是将取值为(-∞,+∞)的数映射到(-1,1)之间，其公式为：
tanh函数在0附近很短一段区域内可看做线性的。由于tanh函数均值为0，因此弥补了sigmoid函数均值为0.5的缺点。
ReLU函数，ReLU函数又称为修正线性单元(Rectified Linear Unit)，是一种分段线性函数，其弥补了sigmoid函数以及tanh函数的梯度消失问题。ReLU函数的公式如下：
ReLU函数的优点：
(1)在输入为正数的时候(对于大多数输入z空间来说)，不存在梯度消失问题。
(2)计算速度要快很多。ReLU函数只有线性关系，不管是前向传播还是反向传播，都比sigmod和tanh要快很多。
ReLU函数的缺点：
(1)当输入为负时，梯度为0，会产生梯度消失问题。
在本领域技术人员都能够理解上述基本概念及常规操作方式的基础上，本发明一种基于植株萎蔫程度进行设施蔬菜精准灌溉的系统，所述系统包括：
图像采集模块，通过图像传感器进行植株的图像信息采集，所采集的信息包括：叶片信息及茎部信息；
图像预处理模块，所述图像预处理模块用于完成图像的预处理，包括图像裁剪、分割、增强中的至少一项；
需水量预测模块，将所采集的图像信息输入至经训练的深度神经网络进行计算，预测出当前植株萎蔫程度状态下对应的预测需水量；
灌溉控制模块，根据预测出的所述植株的需水量形成控制指令，以进行精准定量的供水。
进一步，可选的，所述预处理还包括：图像信息的筛选，剔除不满足清晰度要求或未能包含至少一片完整叶片的图像；所述图像裁剪包括：将获得的图像信息进行裁剪，获得叶片信息及茎部信息，并建立叶片与该植株茎部之间的对应关系。
进一步，可选的，所述预处理还包括：将RGB空间中的R与B分量进行色彩抑制，将G通道分量进行增强。
进一步，可选的，所述深度神经网络具体包括多区域卷积神经网络模型，所述多区域卷积神经网络模型包括：深度卷积神经网络，用于生成原始叶片的映射特征；多区域置信网络模型，包括多个区域的置信网络模型，用于对所述植株的当前状态生成不同需水量的多个不同置信度值，对多个区域的不同置信度值进行拟合，确定出在不同区域中置信度值都相对较大的置信度值，将其对应的需水量确定为植株的需水量。
进一步，可选的，所述多区域置信网络模型包括多区域池化层和完全连接层，多区域池化层包括多个区域的池化层，池化层个数为1，池化层还可替换成完全连接层；所述池化层用于生成置信度。
对应的，本发明还提出了一种基于植株萎蔫程度进行精准灌溉的方法，所述方法包括：
通过图像传感器进行植株的图像信息采集，所采集的信息包括：叶片信息及茎部信息；
利用图像预处理模块完成图像的预处理，包括图像裁剪、分割、增强中的至少一项；
将所采集的图像信息输入至经训练的深度神经网络进行计算，预测出当前植株萎蔫程度状态下对应的预测需水量；
根据预测出的所述植株的需水量形成控制指令，以进行精准定量的供水。
进一步，可选的，所述预处理还包括：图像信息的筛选，剔除不满足清晰度 要求或未能包含至少一片完整叶片的图像；所述图像裁剪包括：将获得的图像信息进行裁剪，获得叶片信息及茎部信息，并建立叶片与该植株茎部之间的对应关系。
进一步，可选的，所述预处理还包括：将RGB空间中的R与B分量进行色彩抑制，将G通道分量进行增强。
进一步，可选的，所述深度神经网络具体包括多区域卷积神经网络模型，所述多区域卷积神经网络模型包括：深度卷积神经网络，用于生成原始叶片的映射特征；多区域置信网络模型，包括多个区域的置信网络模型，用于对所述植株的当前状态生成不同需水量的多个不同置信度值，对多个区域的不同置信度值进行拟合，确定出在不同区域中置信度值都相对较大的置信度值，将其对应的需水量确定为植株的需水量。
进一步，可选的，所述多区域置信网络模型包括多区域池化层和完全连接层，多区域池化层包括多个区域的池化层，池化层个数为1，池化层还可替换成完全连接层；所述池化层用于生成置信度。
本申请还对应提出了一种计算机存储介质，所述存储介质上存储有程序代码，所述代码用于实现上述任一种所述的方法。
本申请还对应提出了一种计算机设备，所述设备包括处理器、存储器，所述存储器上存储有计算机指令，所述指令用于实现上述任一种所述的方法。
再次陈述本申请的有益效果：
1.本申请利用一种基于植株萎蔫程度进行精准灌溉的系统及方法，实现了植株的精准灌溉控制，针对多个叶片和茎部缺水特征，形成多个置信度判断网络，综合得出植株需水量，防止个别叶片存在特殊病变等情况影响判断的准确性。
2.本申请为了提升需水量的预测精度和训练速度，在池化层、激励函数、损失函数的使用上，都采用了新的算法，以整体上提高深度神经网络的训练的精度和速度。在灌溉控制领域，属于申请人首次提出，因此并非常规技术手段或公知常识。
3.在叶片的预处理上，针对绝大部分植株叶片都是绿色的特点，对于识别到的叶片，增强其G分量，相对抑制其R、B分量，有利于提升需求量预测判断的准确性。
图1表示本申请的基本实施例的结构示意图。
为了使本发明的目的、技术方案和优点更加清楚，下面将结合附图对本发明作进一步地详细描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其它实施例，都属于本发明保护的范围。
如图1所示，本申请提出了一种本发明一种基于植株萎蔫程度进行精准灌溉的系统，所述系统包括：
图像采集模块，通过图像传感器进行植株的图像信息采集，所采集的信息包括：叶片信息及茎部信息；
图像预处理模块，所述图像预处理模块用于完成图像的预处理，包括图像裁剪、分割、增强中的至少一项；
需水量预测模块，将所采集的图像信息输入至经训练的深度神经网络进行计算，预测出当前植株萎蔫程度状态下对应的预测需水量；
灌溉控制模块，根据预测出的所述植株的预测需水量形成控制指令，以进行精准定量的供水。
进一步，可选的，所述预处理还包括：图像信息的筛选，剔除不满足清晰度要求或未能包含至少一片完整叶片的图像；所述图像裁剪包括：将获得的图像信息进行裁剪，获得叶片信息及茎部信息，并建立叶片与该植株茎部之间的对应关系。
进一步，可选的，所述预处理还包括：将RGB空间中的R与B分量进行色彩抑制，将G通道分量进行增强。
进一步，可选的，所述深度神经网络具体包括多区域卷积神经网络模型，所述多区域卷积神经网络模型包括：深度卷积神经网络，用于生成原始叶片的映射特征；多区域置信网络模型，包括多个区域的置信网络模型，用于对所述植株的当前状态生成不同需水量的多个不同置信度值，对多个区域的不同置信度值进行拟合，确定出在不同区域中置信度值都相对较大的置信度值，将其对应的需水量确定为植株的需水量。
进一步，可选的，所述多区域置信网络模型包括多区域池化层和完全连接层，多区域池化层包括多个区域的池化层，池化层个数为1，池化层还可替换成完全连接层；所述池化层用于生成置信度。
进一步，可选的，对G分量进行增强，首先对分割后的叶片进行R、G、B分量的分离；生成G分量的调节系数：
然后利用调节系数进行调整：
利用调整后的G
F逆向合成叶片图像；
其中各分量的分解及合成属于本领域的现有技术，但是上述调节方式是本发明人的首创，经过上述调整，能够充分利用叶片的绿色通道信息，使得后续的预测经过更加准确。
进一步，可选的，所述图像分割采用了改进的分水岭分割方式进行分割：
其中Gradient(x,y)表示像素点(x,y)的原始梯度值；
分别表示在窗口D区域内的梯度均值、梯度最小值、梯度最大值；Gra表示修正梯度值；
S＝watershed(Gra)
进一步，可选的，所述深度神经网络包括深度卷积神经网络及多区域置信网络，所述深度卷积神经网络，用于生成分割出的各叶片及茎部特征；多区域置信网络为包括多个叶片区域以及茎部区域的置信网络模型，用于对多个不同叶片及茎部的缺水程度产生置信度值，当至少两个叶片区域的置信度值以及对应茎部区域的置信度值均满足预设的第一阈值等级范围(对应于叶片)、第二阈值等级范围(对应于茎部)时，则确定植株的需水量。
进一步，可选的，所述深度神经网络包括深度卷积神经网络及多区域置信网络，所述深度卷积神经网络能够接收输入植株的分割后叶片及茎部图像，并生成不同尺度的卷积多特征映射；所述多区域置信网络模型包括多区域池化层和完全连接层；其中，所述多区域池化层包括多个区域池化层；所述多个区域池化层用 于生成植株图像需水量的置信度值，当至少两个叶片区域的置信度值以及对应茎部区域的置信度值均满足预设的第一阈值等级范围(对应于叶片)、第二阈值等级范围(对应于茎部)时，则确定植株的需水量；所述多区域池化层设置为最大池化层或平均池化层；所述完全连接层用于对识别到的叶片及茎部缺水程度进行分类。
进一步，可选的，所述深度神经网络为深度卷积神经网络，具体包括：输入层、嵌入层、池化层、全连接层；所述输入层用于接收输入植株的分割后叶片及茎部图像；所述嵌入层采用的卷积核大小为5*5；本发明的激励函数记为R
l()；经过全连接层处理后进一步得到需水量预测推荐结果；
所述池化层的池化方法如下：
R
l()表示激励函数，w
e表示当前层的权重，φ表示损失函数，x
e-1表示上一层的输出；
激励函数R
l为：
所述损失函数φ如下：
N表示正样本数据集的大小，i取值1～N，y
i表示正样本x
i对应的标签值；W
yi表示正样本特征向量x
i在其标签y
i处的权重，s为深度卷积神经网络的推荐参数；b
j表示样本x
i在其标签y
i处的偏差。
进一步，可选的，茎部的状态可以利用茎流传感器进行测量，替代图像传感器对茎部状态进行识别，可以将径流传感器的测量结果输入至深度卷积神经网络中进行训练以及预测。
对应的，本发明还提出了一种基于植株萎蔫程度进行精准灌溉的方法，所述方法包括：
通过图像传感器进行植株的图像信息采集，所采集的信息包括：叶片信息及茎部信息；
利用图像预处理模块完成图像的预处理，包括图像裁剪、分割、增强中的至少一项；
将所采集的图像信息输入至经训练的深度神经网络进行计算，预测出当前植株萎蔫程度状态下对应的预测需水量；
根据预测出的所述植株的需水量形成控制指令，以进行精准定量的供水。
进一步，可选的，所述预处理还包括：图像信息的筛选，剔除不满足清晰度要求或未能包含至少一片完整叶片的图像；所述图像裁剪包括：将获得的图像信息进行裁剪，获得叶片信息及茎部信息，并建立叶片与该植株茎部之间的对应关系。
进一步，可选的，所述预处理还包括：将RGB空间中的R与B分量进行色彩抑制，将G通道分量进行增强。
进一步，可选的，所述深度神经网络具体包括多区域卷积神经网络模型，所述多区域卷积神经网络模型包括：深度卷积神经网络，用于生成原始叶片的映射特征；多区域置信网络模型，包括多个区域的置信网络模型，用于对所述植株的当前状态生成不同需水量的多个不同置信度值，对多个区域的不同置信度值进行拟合，确定出在不同区域中置信度值都相对较大的置信度值，将其对应的需水量确定为植株的需水量。
进一步，可选的，所述多区域置信网络模型包括多区域池化层和完全连接层，多区域池化层包括多个区域的池化层，池化层个数为1，池化层还可替换成完全连接层；所述池化层用于生成置信度。
进一步，可选的，对G分量进行增强，首先对分割后的叶片进行R、G、B分量的分离；生成G分量的调节系数：
然后利用调节系数进行调整：
利用调整后的G
F逆向合成叶片图像；
其中各分量的分解及合成属于本领域的现有技术，但是上述调节方式是本发明人的首创，经过上述调整，能够充分利用叶片的绿色通道信息，使得后续的预测经过更加准确。
进一步，可选的，所述图像分割采用了改进的分水岭分割方式进行分割：
其中Gradient(x,y)表示像素点(x,y)的原始梯度值；
分别表示在窗口D区域内的梯度均值、梯度最小值、梯度最大值；Gra表示修正梯度值；
S＝watershed(Gra)
进一步，可选的，所述深度神经网络包括深度卷积神经网络及多区域置信网络，所述深度卷积神经网络，用于生成分割出的各叶片及茎部特征；多区域置信网络为包括多个叶片区域以及茎部区域的置信网络模型，用于对多个不同叶片及茎部的缺水程度产生置信度值，当至少两个叶片区域的置信度值以及对应茎部区域的置信度值均满足预设的第一阈值等级范围(对应于叶片)、第二阈值等级范围(对应于茎部)时，则确定植株的需水量。
进一步，可选的，所述深度神经网络包括深度卷积神经网络及多区域置信网络，所述深度卷积神经网络能够接收输入植株的分割后叶片及茎部图像，并生成不同尺度的卷积多特征映射；所述多区域置信网络模型包括多区域池化层和完全连接层；其中，所述多区域池化层包括多个区域池化层；所述多个区域池化层用于生成植株图像需水量的置信度值，当至少两个叶片区域的置信度值以及对应茎部区域的置信度值均满足预设的第一阈值等级范围(对应于叶片)、第二阈值等级范围(对应于茎部)时，则确定植株的需水量；所述多区域池化层设置为最大池化层或平均池化层；所述完全连接层用于对识别到的叶片及茎部缺水程度进行分类。
进一步，可选的，所述深度神经网络为深度卷积神经网络，具体包括：输入层、嵌入层、池化层、全连接层；所述输入层用于接收输入植株的分割后叶片及茎部图像；所述嵌入层采用的卷积核大小为5*5；本发明的激励函数记为R
l()；经过全连接层处理后进一步得到需水量预测推荐结果；
所述池化层的池化方法如下：
R
l()表示激励函数，w
e表示当前层的权重，φ表示损失函数，x
e-1表示上一层的输出；
激励函数R
l为：
所述损失函数φ如下：
N表示正样本数据集的大小，i取值1～N，y
i表示正样本x
i对应的标签值；W
yi表示正样本特征向量x
i在其标签y
i处的权重，s为深度卷积神经网络的推荐参数；b
j表示样本x
i在其标签y
i处的偏差。
进一步，可选的，茎部的状态可以利用茎流传感器进行测量，替代图像传感器对茎部状态进行识别，可以将径流传感器的测量结果输入至深度卷积神经网络中进行训练以及预测。本申请还对应提出了一种计算机存储介质，所述存储介质上存储有程序代码，所述代码用于实现上述任一种所述的方法。
本申请还对应提出了一种计算机设备，所述设备包括处理器、存储器，所述存储器上存储有计算机指令，所述指令用于实现上述任一种所述的方法。
在本说明书的描述中，参考术语“一个实施例”、“示例”、“具体示例”等的描述意指结合该实施例或示例描述的具体特征、结构、材料或者特点包含于本发明的至少一个实施例或示例中。在本说明书中，对上述术语的示意性表述不一定指的是相同的实施例或示例。
本申请还提出一种计算机可读介质，上面包含可实现上述系统的程序代码，所包含的程序代码可以用任何适当的介质传输，包括但不限于无线、电线、光缆、 RF等等，或者上述的任意合适的组合。计算机可读介质可以是计算机可读信号介质或者计算机可读存储介质。计算机可读存储介质例如可以是电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。计算机可读存储介质的更具体的例子(非穷举的列表)包括：具有一个或多个导线的电连接、便携式计算机磁盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、光纤、便携式紧凑磁盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。在本文件中，计算机可读存储介质可以是任何包含或存储程序的有形介质，该程序可以被指令执行系统、装置或者器件使用或者与其结合使用。可以以一种或多种程序设计语言或其组合来编写用于执行本发明操作的计算机程序代码，所述程序设计语言包括面向对象的程序设计语言—诸如Java、Smalltalk、C++，还包括常规的过程式程序设计语言—诸如“C”语言或类似的程序设计语言。程序代码可以完全地在用户计算机上执行、部分地在用户计算机上执行、作为一个独立的软件包执行、部分在用户计算机上部分在远程计算机上执行、或者完全在远程计算机或服务器上执行。在涉及远程计算机的情形中，远程计算机可以通过任意种类的网络——包括局域网(LAN)或广域网(WAN)连接到用户计算机，或者，可以连接到外部计算机(例如利用因特网服务提供商来通过因特网连接)。上述以软件功能单元的形式实现的集成的单元，可以存储在一个计算机可读取存储介质中。上述软件功能单元存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)或处理器(processor)执行本发明各个实施例所述方法的部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(Read-Only Memory，ROM)、随机存取存储器(Random Access Memory，RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述仅为本发明的优选实施例，并非因此限制本发明的专利范围，凡是利用本发明说明书及附图内容所作的等效结构或等效流程变换，直接或间接运用在其他相关的技术领域，均同理包括在本发明的专利保护范围内。以上公开的本发明优选实施例只是用于帮助阐述本发明。优选实施例并没有详尽叙述所有的细节，也不限制该发明仅为所述的具体实施方式。显然，根据本说明书的内容，可作很多的修改和变化。本说明书选取并具体描述这些实施例，是为了更好地解释 本发明的原理和实际应用，从而使所属技术领域技术人员能很好地理解和利用本发明。本发明仅受权利要求书及其全部范围和等效物的限制。
Claims (6)
- 一种基于植株萎蔫程度进行精准灌溉的系统，所述系统包括：图像采集模块，通过图像传感器进行植株的图像信息采集，所采集的信息包括：叶片信息及茎部信息；图像预处理模块，所述图像预处理模块用于完成图像的预处理，包括图像裁剪、分割、增强中的至少一项；需水量预测模块，将所采集的图像信息输入至经训练的深度神经网络进行计算，预测出当前植株萎蔫程度状态下对应的预测需水量；灌溉控制模块，根据预测出的所述植株的需水量形成控制指令，以进行定时定量的供水；所述深度神经网络具体包括多区域卷积神经网络模型，所述多区域卷积神经网络模型包括：深度卷积神经网络，用于生成原始叶片的映射特征；多区域置信网络模型，包括多个区域的置信网络模型，用于对所述植株的当前状态生成不同需水量的多个不同置信度值，对多个区域的不同置信度值进行拟合，确定出在不同区域中置信度值都相对较大的置信度值，将其对应的需水量确定为植株的需水量；所述多区域置信网络模型包括多区域池化层和完全连接层，多区域池化层包括多个区域的池化层，池化层个数为1；所述池化层用于生成置信度。
- 根据权利要求1所述的一种基于植株萎蔫程度进行精准灌溉的系统，所述预处理还包括：图像信息的筛选，剔除不满足清晰度要求或未能包含至少一片完整叶片的图像；所述图像裁剪包括：将获得的图像信息进行裁剪，获得叶片信息及茎部信息，并建立叶片与该植株茎部之间的对应关系。
- 根据权利要求1所述的一种基于植株萎蔫程度进行精准灌溉的系统，所述预处理还包括：将RGB空间中的R与B分量进行色彩抑制，将G通道分量进行增强。
- 一种基于植株萎蔫程度进行精准灌溉的方法，所述方法包括：通过图像传感器进行植株的图像信息采集，所采集的信息包括：叶片信息及茎部信息；利用图像预处理模块完成图像的预处理，包括图像裁剪、分割、增强中的至少一项；将所采集的图像信息输入至经训练的深度神经网络进行计算，预测出当前植株萎蔫程度状态下对应的预测需水量；根据预测出的所述植株的需水量形成控制指令，以进行定时定量的供水；所述深度神经网络具体包括多区域卷积神经网络模型，所述多区域卷积神经网络模型包括：深度卷积神经网络，用于生成原始叶片的映射特征；多区域置信网络模型，包括多个区域的置信网络模型，用于对所述植株的当前状态生成不同需水量的多个不同置信度值，对多个区域的不同置信度值进行拟合，确定出在不同区域中置信度值都相对较大的置信度值，将其对 应的需水量确定为植株的需水量；所述多区域置信网络模型包括多区域池化层和完全连接层，多区域池化层包括多个区域的池化层，池化层个数为1；所述池化层用于生成置信度。
- 根据权利要求4所述的一种基于植株萎蔫程度进行精准灌溉的方法，所述预处理还包括：图像信息的筛选，剔除不满足清晰度要求或未能包含至少一片完整叶片的图像；所述图像裁剪包括：将获得的图像信息进行裁剪，获得叶片信息及茎部信息，并建立叶片与该植株茎部之间的对应关系。
- 根据权利要求4所述的一种基于植株萎蔫程度进行精准灌溉的方法，所述预处理还包括：将RGB空间中的R与B分量进行色彩抑制，将G通道分量进行增强。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110971230.2 | 2021-08-23 | ||
| CN202110971230.2A CN113545280B (zh) | 2021-08-23 | 2021-08-23 | 一种基于植株萎蔫程度进行精准灌溉的系统及方法 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2023024615A1 true WO2023024615A1 (zh) | 2023-03-02 |
Family
ID=78105981
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2022/094812 WO2023024615A1 (zh) | 2021-08-23 | 2022-05-25 | 一种基于植株萎蔫程度进行精准灌溉的系统及方法 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN113545280B (zh) |
| WO (1) | WO2023024615A1 (zh) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118470578A (zh) * | 2024-07-11 | 2024-08-09 | 泰安市园林绿化管理服务中心 | 一种园林精细化管理与自动检测方法 |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113545280B (zh) * | 2021-08-23 | 2022-01-25 | 中国农业科学院蔬菜花卉研究所 | 一种基于植株萎蔫程度进行精准灌溉的系统及方法 |
| CN116784217A (zh) * | 2023-06-15 | 2023-09-22 | 四川省农业机械科学研究院 | 基于风能、太阳能和土壤湿度的智慧农业灌溉系统 |
| CN117079060B (zh) * | 2023-10-13 | 2024-03-12 | 之江实验室 | 一种基于光合信号的叶片智能分类方法和系统 |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102524024A (zh) * | 2012-02-16 | 2012-07-04 | 四川农业大学 | 基于计算机视觉的作物灌溉系统 |
| CN110178518A (zh) * | 2019-07-03 | 2019-08-30 | 福州阿里他巴信息科技有限公司 | 一种水肥灌溉系统 |
| CN112514779A (zh) * | 2020-12-09 | 2021-03-19 | 广州大学华软软件学院 | 一种智能灌溉系统 |
| US20210204496A1 (en) * | 2020-01-08 | 2021-07-08 | The United States Of America, As Represented By The Secretary Of Agriculture | System and method of watering crops with a variable rate irrigation system |
| CN113545280A (zh) * | 2021-08-23 | 2021-10-26 | 中国农业科学院蔬菜花卉研究所 | 一种基于植株萎蔫程度进行精准灌溉的系统及方法 |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101411299B (zh) * | 2008-11-25 | 2010-08-25 | 浙江大学 | 植物叶片膨压感应节水灌溉自动控制方法 |
| CN108199954B (zh) * | 2018-01-23 | 2019-01-08 | 南京万荣立体绿化工程有限公司 | 灌溉智能监测系统 |
| CN110702882A (zh) * | 2019-10-28 | 2020-01-17 | 沈阳农业大学 | 一种基于作物水分分级预警的动态决策系统 |
| CN112119887A (zh) * | 2020-10-27 | 2020-12-25 | 博尔塔拉蒙古自治州飒博尔农业科技有限公司 | 一种智能阀门灌溉的控制装置及其控制方法 |
| CN112330694A (zh) * | 2020-11-16 | 2021-02-05 | 新疆农业科学院粮食作物研究所 | 一种植物萎蔫程度计算方法及系统 |
-
2021
- 2021-08-23 CN CN202110971230.2A patent/CN113545280B/zh active Active
-
2022
- 2022-05-25 WO PCT/CN2022/094812 patent/WO2023024615A1/zh active Application Filing
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102524024A (zh) * | 2012-02-16 | 2012-07-04 | 四川农业大学 | 基于计算机视觉的作物灌溉系统 |
| CN110178518A (zh) * | 2019-07-03 | 2019-08-30 | 福州阿里他巴信息科技有限公司 | 一种水肥灌溉系统 |
| US20210204496A1 (en) * | 2020-01-08 | 2021-07-08 | The United States Of America, As Represented By The Secretary Of Agriculture | System and method of watering crops with a variable rate irrigation system |
| CN112514779A (zh) * | 2020-12-09 | 2021-03-19 | 广州大学华软软件学院 | 一种智能灌溉系统 |
| CN113545280A (zh) * | 2021-08-23 | 2021-10-26 | 中国农业科学院蔬菜花卉研究所 | 一种基于植株萎蔫程度进行精准灌溉的系统及方法 |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118470578A (zh) * | 2024-07-11 | 2024-08-09 | 泰安市园林绿化管理服务中心 | 一种园林精细化管理与自动检测方法 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113545280B (zh) | 2022-01-25 |
| CN113545280A (zh) | 2021-10-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2022253057A1 (zh) | 一种用于日光温室果菜栽培的智能水分精准灌溉控制系统及方法 | |
| WO2023024615A1 (zh) | 一种基于植株萎蔫程度进行精准灌溉的系统及方法 | |
| Khairunniza-Bejo et al. | Application of artificial neural network in predicting crop yield: A review | |
| US20220075344A1 (en) | A method of finding a target environment suitable for growth of a plant variety | |
| CN117036088A (zh) | 一种ai识别绿化植物生长态势的数据采集分析方法 | |
| CN110084417A (zh) | 一种基于grnn神经网络的草莓温室环境参数智能监测系统 | |
| CN109978234A (zh) | 墒情预测、灌溉决策方法和装置 | |
| CN102789579A (zh) | 基于计算机视觉技术的温室作物水肥胁迫状态识别方法 | |
| CN110119169A (zh) | 一种基于最小向量机的番茄温室温度智能预警系统 | |
| Tariqul Islam et al. | Automatic detection of Grape, Potato and Strawberry Leaf Diseases using CNN and image processing | |
| CN115115830A (zh) | 一种基于改进Transformer的家畜图像实例分割方法 | |
| Barriga et al. | Crop-water assessment in Citrus (Citrus sinensis L.) based on continuous measurements of leaf-turgor pressure using machine learning and IoT | |
| Gao et al. | How can agricultural water production be promoted? A review on machine learning for irrigation | |
| Li et al. | Low-carbon jujube moisture content detection based on spectral selection and reconstruction | |
| Niu et al. | YOLO-plum: A high precision and real-time improved algorithm for plum recognition | |
| Mezouari et al. | A Hadoop based framework for soil parameters prediction | |
| Monica et al. | Soil NPK prediction using enhanced genetic algorithm | |
| Hao et al. | Automatic acquisition, analysis and wilting measurement of cotton 3D phenotype based on point cloud | |
| CN116957141A (zh) | 一种基于Transformer模型的智能灌溉预测方法及系统 | |
| CN116579873A (zh) | 基于高温干旱气象评估作物灾害减产情况的方法及系统 | |
| JP7452879B2 (ja) | 予測システム、予測方法、および予測プログラム | |
| Dong et al. | Research on graded diagnosis of lettuce water-nitrogen stress and pest prevention based on deep learning | |
| Jimenez Lopez et al. | Forecasting irrigation scheduling based on deep learning models using IoT | |
| Mayuri et al. | Artificial Neural Network (ANN) With Chan-Vese (CV) Algorithm-Based Plant Disease Detection And Classification | |
| TW202025063A (zh) | 作物生長階段計算方法及電腦程式產品 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 22859963 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 22859963 Country of ref document: EP Kind code of ref document: A1 |