CN113935972A - Method and device for detecting root system image of plant in situ based on micro root canal and storage medium - Google Patents
Method and device for detecting root system image of plant in situ based on micro root canal and storage medium Download PDFInfo
- Publication number
- CN113935972A CN113935972A CN202111210073.XA CN202111210073A CN113935972A CN 113935972 A CN113935972 A CN 113935972A CN 202111210073 A CN202111210073 A CN 202111210073A CN 113935972 A CN113935972 A CN 113935972A
- Authority
- CN
- China
- Prior art keywords
- image
- root
- plant
- root system
- micro
- 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
- 238000000034 method Methods 0.000 title claims abstract description 49
- 210000004262 dental pulp cavity Anatomy 0.000 title claims abstract description 28
- 238000011065 in-situ storage Methods 0.000 title claims abstract description 22
- 238000003860 storage Methods 0.000 title claims abstract description 17
- 241000196324 Embryophyta Species 0.000 claims abstract description 74
- 238000001514 detection method Methods 0.000 claims abstract description 23
- 230000011218 segmentation Effects 0.000 claims abstract description 18
- 230000012010 growth Effects 0.000 claims abstract description 9
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 238000012549 training Methods 0.000 claims description 36
- 238000011176 pooling Methods 0.000 claims description 28
- 230000006870 function Effects 0.000 claims description 22
- 230000004913 activation Effects 0.000 claims description 18
- 230000007246 mechanism Effects 0.000 claims description 18
- 238000004590 computer program Methods 0.000 claims description 15
- 238000003709 image segmentation Methods 0.000 claims description 13
- 239000002689 soil Substances 0.000 claims description 10
- 238000012360 testing method Methods 0.000 claims description 9
- 230000008635 plant growth Effects 0.000 claims description 8
- 238000000605 extraction Methods 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 7
- 238000013508 migration Methods 0.000 claims description 6
- 230000005012 migration Effects 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 5
- 238000013135 deep learning Methods 0.000 claims description 5
- 238000013528 artificial neural network Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 238000013526 transfer learning Methods 0.000 claims description 4
- 241000512668 Eunectes Species 0.000 claims description 3
- 230000009191 jumping Effects 0.000 claims description 3
- 238000007789 sealing Methods 0.000 claims description 3
- 238000013210 evaluation model Methods 0.000 claims description 2
- 239000000463 material Substances 0.000 claims description 2
- 230000002786 root growth Effects 0.000 claims description 2
- 241000238631 Hexapoda Species 0.000 abstract description 3
- 241000607479 Yersinia pestis Species 0.000 abstract description 3
- 201000010099 disease Diseases 0.000 abstract description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 abstract description 3
- 238000010561 standard procedure Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 9
- 238000012544 monitoring process Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- NIXOWILDQLNWCW-UHFFFAOYSA-N acrylic acid group Chemical group C(C=C)(=O)O NIXOWILDQLNWCW-UHFFFAOYSA-N 0.000 description 2
- 230000001066 destructive effect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 239000002390 adhesive tape Substances 0.000 description 1
- 238000009412 basement excavation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000007689 inspection 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
- 235000015097 nutrients Nutrition 0.000 description 1
- 210000000056 organ Anatomy 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 108700002776 rat Trim25 Proteins 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10056—Microscopic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
A method, equipment and a storage medium for detecting an image of a root system in situ of a plant based on a micro root canal belong to the field of image detection methods, and are provided for solving the problems that a standard method for tracking the root in an image is extremely complicated and time-consuming, and the accuracy of manual detection is low, and the method comprises the following steps: (1) carrying out image acquisition on a plant root system; (2) performing semantic annotation on the acquired image; (3) preprocessing an original image and a corresponding label file; (4) a target segmentation system platform. The artificial dependence is low, the comprehensive cost is low, the efficiency is high, the precision is high, the growth state of the plant is comprehensively recorded, and the problem of plant diseases and insect pests of the plant root system can be found in time.
Description
Technical Field
The invention relates to the field of image detection methods, in particular to a method, equipment and a storage medium for detecting a root system image of a plant in situ based on micro root canals.
Background
In plant growth, the root system is an important organ for plants to obtain nutrients and water from medium environments such as soil. Because of the special requirements of the root system on the plant growth, the growth condition of the plant root system under different conditions is known, and the method has very important significance in the fields of the plant, the agriculture, the crop cultivation and the like. But the plant roots are generally located in the soil and are difficult to detect,
at present, destructive observation methods such as an earth boring method, an excavation method and the like and non-destructive observation methods such as a micro root canal method, a container method and the like are mainly adopted for observing plant root systems. The conventional method has the disadvantages of high manual dependence, low efficiency and low accuracy.
In the absence of computer vision: the method is characterized in that a micro root canal is used for collecting a plant picture, a micro root window image collecting system collects root system images in the deep soil, the collecting environment of the images is poor, the images become unclear and the image contrast is poor due to factors such as light diffraction, uneven illumination, poor camera focusing and the like, the collected image background contains a plurality of soil particles and gravels with gray values similar to the root system, great difficulty is brought to image segmentation, and the standard method for tracking the roots in the images is extremely complex and time-consuming.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention provides the method, the equipment and the storage medium for detecting the in-situ plant root system image based on the micro root canal, which can solve the problems of extremely complicated and time-consuming standard method for tracking the root in the image and low accuracy, greatly improve the working efficiency, reduce the labor cost and ensure the identification accuracy.
The technical scheme adopted by the invention is as follows:
the method for detecting the in-situ root system image of the plant based on the micro root canal comprises the following steps:
s1, acquiring images of the plant root system;
step S2, carrying out semantic annotation on the collected image;
step S3, preprocessing the original image and the corresponding label file;
step S4 is to perform image segmentation detection on the data processed in step S3.
Further, in step S1, the implementation of image acquisition on the plant root system is as follows:
step S11, setting up a micro-root window tube detection device, processing the micro-root window tube detection device into a micro-root tube observation shell by using a tube, covering two ends of the tube with lightproof black covers, sealing the part exposed on the soil by using lightproof materials to prevent light from entering the tube to influence plant growth, embedding the tube into the soil near the plant, and waiting for the growth of a plant root system;
step S12, acquiring data, regularly detecting a plant root growth picture by using an endoscope, and acquiring a plant root data set;
and step S13, marking and sorting.
Further, in step S2, the semantic annotation performed on the acquired image is implemented as follows:
LabelMe is installed in Anaconda, a graphic image annotation tool LabelmE supporting semantic segmentation is used for annotating the image in a folder, automatic saving is set, the root part in the original image is annotated and then is saved as a json file, the json file is converted into a gray icon label file through a program, the root area is given with a pixel value of 1, and the background area is given with a pixel value of 0.
Further, in step S3, the implementation manner of preprocessing the original image and the corresponding tag file is as follows:
the image is uniformly divided into 512 x 512, and the random rotation, horizontal turning and mirror image turning are carried out on the image, noise is randomly added, and data samples are increased.
Further, in step S4, the image segmentation detection method is as follows:
building a plant root system image segmentation neural network, training a segmentation network model, adjusting parameters, dividing an obtained image into a training image and a test image according to a ratio of 8:2, training the network by using the training image and a corresponding label file thereof in a model training stage, and obtaining trained weight parameters to segment the test image;
and training the deep learning segmentation network model by using a sample set similar to the root system to obtain corresponding training parameters.
Further, the split network model is implemented as follows:
building a U-net network combined with a transfer learning and attention mechanism:
setting a main stem extraction part in the U-net as a VGG16 network for extracting the characteristics of the image;
firstly, in the process of extracting a main trunk of an image, performing convolution and Relu activation functions of 64 channels twice on a preprocessed image with the size of 512 × 3 by using VGG16, wherein 512 × 512 is the image scale, 3 is the number of channels of the image, a primary effective feature layer C1 is obtained, then performing 2 × 2 maximum pooling, wherein the maximum pooling does not change the number of channels of the image, and the maximum pooling compresses the height and width of the image to half to obtain 256 × 64 images;
performing two 128-channel convolution and Relu activation functions on the obtained 256 × 64 images to obtain an effective feature layer C2, and continuing to perform 2 × 2 maximal pooling, wherein the maximal pooling compresses the height and width of the images to half to obtain 128 × 128 images;
thirdly, performing 256-channel convolution and Relu activation functions on the obtained 128 × 128 images for three times to obtain an effective feature layer C3, and continuing to perform 2 × 2 maximum pooling, wherein the maximum pooling can compress the height and width of the images to half to obtain 64 × 256 images;
and fourthly, performing convolution of the obtained 64 x 256 images by three 512 channels and a Relu activation function to obtain an effective feature layer C4, and continuing to perform maximum pooling of 2 x 2, wherein the maximum pooling compresses the height and width of the images to half to obtain 32 x 512 images.
Fifthly, performing convolution with the step length of 1 x 1 and the channel of 512 on the image obtained in the step (iv) for three times to obtain a feature layer K of 32 x 512;
adding an attention mechanism into the U-net upsampling, overlapping the K and an effective feature layer C4 connected in a jumping mode with the upsampled K after the K passes through the attention mechanism, generating K1 after two 3 x 3 convolutions and Relu activation functions, overlapping the K1 and the C3 with the upsampled K1 after passing through the attention mechanism, generating K2 after the same two 3 x 3 convolutions and the Relu activation functions, repeating the steps twice, and obtaining a feature map with the same size as the input image through one 1 x 1 convolution;
migrating model parameters obtained by a sample set similar to a root system by using migration learning, inputting training data into a U-net network combined with the migration learning and attention mechanism for parameter training, and storing the parameters obtained by training; and finally, segmenting the test data set by using the trained weight to obtain a segmentation result graph.
Further, the method also comprises a prediction network, and the implementation mode is as follows:
detecting detection data by using a model file obtained by a U-net network combining a training attention mechanism to realize prediction of root system segmentation;
calculating an AUC value through an ROC curve in each training period, and judging the quality of the model by using the AUC value, wherein the method is specifically realized as follows:
formula for false positive rate FPR:
equation of true positive ratio TPR:
in the formula (1), FPR is the horizontal axis of the ROC characteristic curve; FP is originally a non-root sample, but is identified by the network as a root sample; n is all non-root samples;
in the formula (2), TPR is the vertical axis of the ROC characteristic curve; TP is identifying the correct root sample; p is all root samples;
the larger the TPR is, the higher the proportion of the positive samples which are correctly identified is, and the smaller the FPR is, the smaller the proportion of the negative samples which are wrongly identified as the positive samples is;
the AUC is the area under the ROC curve, is between 0.1 and 1, and is used as the quality of a numerical evaluation model, and the larger the value is, the better the model is represented.
The device for detecting the original position root system image of the plant based on the micro root canal comprises a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to realize the steps of the method for detecting the original position root system image of the plant based on the micro root canal.
A computer-readable storage medium having stored thereon a computer program, characterized in that: the computer program when executed by a processor implements the method for detecting the root system image of the micro root canal based plant in situ as described in any one of the above.
The invention has the beneficial effects that:
1. the labor dependence is low, and the labor cost, the time cost and the like are greatly reduced.
2. The working efficiency is high, and the workload can be greatly increased in the same time.
3. The recognition accuracy is high, and the working quality is high.
4. The growth state of the plant is comprehensively recorded, and the problem of plant diseases and insect pests of the plant root system can be found in time.
Drawings
FIG. 1 is a schematic block diagram of a method for detecting an in situ root system image of a plant based on micro-root canals;
FIG. 2 is a flow chart of a system implementation of a method for in situ root imaging inspection of a plant based on micro-canals;
FIG. 3 is a schematic diagram of image acquisition of a method for detecting an in-situ root system image of a plant based on micro-root canals;
FIG. 4 is a schematic diagram of a U-net network in a method for detecting root system images in situ of a plant based on micro root canals;
Detailed Description
The embodiments of the present invention are described below with reference to specific embodiments, and other advantages and effects of the present invention will be easily understood by those skilled in the art from the disclosure of the present specification. The invention is capable of other and different embodiments and of being practiced or of being carried out in various ways, and its several details are capable of modification in various respects, all without departing from the spirit and scope of the present invention. It is to be noted that the features in the following embodiments and examples may be combined with each other without conflict.
The method, the equipment and the storage medium for detecting the root system image of the plant in situ based on the micro root canal have the following application scenes:
1. in the aspect of scientific research:
(1) application scenarios: plant research in natural ecological environment;
(2) the application method comprises the following steps: the image acquisition equipment is arranged near the plant, and regular interval data acquisition and detection are carried out on the root system of the plant.
(3) The application effect is as follows:
the change of plant root systems and the growing environment can be observed by human eyes;
and secondly, providing sufficient data for basic research in plant science fields such as the growth state and the growth environment of the target plant in the monitoring range.
2. And (4) safety monitoring:
(1) application scenarios: planting precious plants;
(2) the application method comprises the following steps: the plant growth state monitoring is combined with the system to carry out the track monitoring of the plant growth state condition.
(3) The application effect is as follows:
firstly, the growth state of the plants can be perceived in time;
and secondly, preventing plant diseases and insect pests at the first time.
The method, the equipment and the storage medium for detecting the root system image of the plant in situ based on the micro root canal are implemented as follows:
provides an effective plant root system research method based on micro root canals. Through the image that uses little root canal technique normal position harmless collection plant roots, classify the label setting to the image again, through carrying out the target detection to the image that has classification label, give the target frame and detect pseudo-label, realize the training of weak supervision, obtain the plant roots of higher performance and detect, can effectively improve the precision that plant roots surveyed.
A migration learning mechanism is proposed for the problem of few data sets. The method of transfer learning is utilized to apply the knowledge obtained by solving one problem to another different but related problem to reuse the knowledge, thereby improving the image segmentation quality and solving the dilemma of a small amount of labeled data in the real situation.
The single U-net model is not accurate enough when applied to feature extraction which is easy to generate in image segmentation, an attention mechanism is added to enable features extracted in the U-net to be more definite, the detection accuracy of the plant root system is improved, and a root system image segmentation system based on deep learning is established.
And (4) combining the transfer learning and attention mechanism with the U-net to optimize the U-net network for image segmentation detection. The condition control is combined with the neural network to establish a plant root system state detection system, and the detection system specifically comprises the steps of root system image acquisition, image processing and result display.
As shown in fig. 3, a plurality of plants are selected to construct a micro root canal detection system, and image acquisition is carried out on plant roots at different depths and different angles. A root system database is established through a large number of sampling, the root system part is labeled, and deep learning training is carried out on the root system part by using different target segmentation algorithms. And optimizing the root system segmentation system. And a software and hardware environment is built, and a reliable system capable of detecting different plant roots is built.
As shown in fig. 1 and fig. 2, the flow of the embodiment is as follows:
(1) image acquisition of plant roots
And (4) building micro-root window tube detection equipment, acquiring data, and labeling and arranging.
Selecting an acrylic pipe as a micro root canal observation shell, wherein the diameter of the pipe is selected from 4cm and 7 cm;
covering two ends of the acrylic tube with light-tight black covers, sealing the part exposed on the soil with light-tight adhesive tapes to prevent light from entering the tube to influence plant growth, embedding the tube into the soil near the plant, and waiting for the growth of plant roots;
and thirdly, regularly detecting the growth picture of the plant root system by using an endoscope to obtain a plant root system data set.
(2) Semantically annotating acquired images
LabelMe is installed in Anaconda, an image in a folder is annotated by using a graphic image annotation tool LabelMe supporting semantic segmentation, automatic saving is set, and a root system part in an original image is annotated and saved as a json file. And converted into a gray icon label file through a program, wherein the root area is assigned with a pixel value of 1, and the background area is assigned with a pixel value of 0.
(3) Preprocessing an original image and a corresponding tag file
The images are uniformly divided into 512 × 512. Through carrying out random rotation, horizontal upset and mirror image upset to the picture, add the noise at random, increase the data sample, data are more, can effectual reduction model overfitt, strengthen the generalization ability of model.
(4) Object segmentation system platform
The method comprises the steps of building a plant root system image segmentation neural network, training a model, adjusting parameters, reducing errors and improving image segmentation accuracy. And dividing the acquired image into a training image and a test image according to the ratio of 8: 2. When the model is trained, the training images and the corresponding label files are used for training the network, and trained weight parameters are obtained to segment the test images.
And training the deep learning segmentation network model by using a sample set similar to the root system to obtain corresponding training parameters.
As shown in fig. 4, a network of improved U-nets is built:
the U-net network is divided into feature extraction, up-sampling and jumping connection, the feature extraction part is composed of convolution and maximum pooling, the whole framework structure of the part is similar to that of the VGG, and therefore the main extraction part in the U-net is set to be the VGG16 network to be used for feature extraction of the image.
Firstly, in the process of extracting a main trunk of an image, performing twice 64-channel convolution and Relu activation functions on a preprocessed image with the size of 512 × 3 by using VGG16, wherein 512 × 512 is the image scale, 3 is the number of channels of the image, a primary effective feature layer C1 is obtained, then performing 2 × 2 maximum pooling, wherein the maximum pooling does not change the number of channels of the image, and the maximum pooling compresses the height and width of the image to half to obtain 256 × 64 images;
performing two 128-channel convolution and Relu activation functions on the obtained 256 × 64 images to obtain an effective feature layer C2, and continuing to perform 2 × 2 maximal pooling, wherein the maximal pooling compresses the height and width of the images to half to obtain 128 × 128 images;
thirdly, performing 256-channel convolution and Relu activation functions on the obtained 128 × 128 images for three times to obtain an effective feature layer C3, and continuing to perform 2 × 2 maximum pooling, wherein the maximum pooling can compress the height and width of the images to half to obtain 64 × 256 images;
and fourthly, performing convolution of the obtained 64 x 256 images by three 512 channels and a Relu activation function to obtain an effective feature layer C4, and continuing to perform maximum pooling of 2 x 2, wherein the maximum pooling compresses the height and width of the images to half to obtain 32 x 512 images. And fifthly, performing convolution on the image with the step size of 1 × 1 and the channel of 512 three times to obtain a feature layer K with the step size of 32 × 512.
An Attention mechanism is added into the upsampling of U-net, and K and a jump-connected effective characteristic layer C4 are superposed with the upsampled K after passing through the Attention mechanism. K1 was generated after two 3 x 3 convolutions and the Relu activation function. K1 and C3 are superposed with the K1 after up-sampling after an attention mechanism. The same two 3 x 3 convolutions and Relu activation function yields K2. The steps are repeated twice, and finally, a feature map with the same size as the input image is obtained through convolution of 1 x 1.
Model parameters obtained by a sample set similar to a root system are migrated by utilizing migration learning, training data are input into an improved U-net-based network for parameter training, and the trained parameters are stored; and finally, segmenting the test data set by using the trained weight to obtain a segmentation result graph.
And detecting the detection data by taking the obtained model as a pre-training network model in the prediction network so as to realize the prediction of root system segmentation. And calculating an AUC value by the ROC curve in each training period, and judging the quality of the model by using the AUC value.
The ROC characteristic curve is a curve with the horizontal axis of false positive rate FPR and the vertical axis of true positive rate TPR.
False positive rateFP, samples that are not root samples but are identified by the network as roots. N: all non-root samples.
When the TPR is larger. The higher the proportion of positive samples that are correctly identified and the smaller the FPR, the smaller the proportion of negative samples that are misidentified as positive samples. The AUC is the area under the ROC curve, is between 0.1 and 1, and can be used as a numerical value for visually evaluating the quality of the model, wherein the larger the value is, the better the value is.
The present embodiments may be provided as a method, system, or computer program product by those skilled in the art using the systems and methods mentioned in the foregoing embodiments. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects, or a combination of both. Furthermore, the present embodiments may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
A flowchart or block diagram of a method, apparatus (system), and computer program product according to the present embodiments is depicted. It will be understood that each flow or block of the flowchart illustrations or block diagrams, and combinations of flows or blocks in the flowchart illustrations or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows, or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows or block diagram block or blocks.
In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). The memory is an example of a computer-readable medium.
Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.
Claims (9)
1. The method for detecting the root system image of the plant in situ based on the micro root canal is characterized by comprising the following steps:
s1, acquiring images of the plant root system;
step S2, carrying out semantic annotation on the collected image;
step S3, preprocessing the original image and the corresponding label file;
step S4 is to perform image segmentation detection on the data processed in step S3.
2. The method for detecting the in-situ root system image of the plant based on the micro root canal of the claim 1, wherein the step S1 is implemented by acquiring the image of the root system of the plant as follows:
step S11, setting up a micro-root window tube detection device, processing the micro-root window tube detection device into a micro-root tube observation shell by using a tube, covering two ends of the tube with lightproof black covers, sealing the part exposed on the soil by using lightproof materials to prevent light from entering the tube to influence plant growth, embedding the tube into the soil near the plant, and waiting for the growth of a plant root system;
step S12, acquiring data, regularly detecting a plant root growth picture by using an endoscope, and acquiring a plant root data set;
and step S13, marking and sorting.
3. The method for detecting the in-situ root system image of the plant based on the micro root canal as claimed in claim 1, wherein the step S2 is implemented by performing semantic annotation on the acquired image as follows:
LabelMe is installed in Anaconda, a graphic image annotation tool LabelmE supporting semantic segmentation is used for annotating the image in a folder, automatic saving is set, the root part in the original image is annotated and then is saved as a json file, the json file is converted into a gray icon label file through a program, the root area is given with a pixel value of 1, and the background area is given with a pixel value of 0.
4. The method for detecting the original position root system image of the plant based on the micro root canal as claimed in claim 1, wherein the step S3 is implemented by preprocessing the original image and the corresponding label file as follows:
the image is uniformly divided into 512 x 512, and the random rotation, horizontal turning and mirror image turning are carried out on the image, noise is randomly added, and data samples are increased.
5. The method for detecting the in-situ root system image of the plant based on the micro root canal of the claim 1, wherein the step S4 is implemented by performing image segmentation detection as follows:
building a plant root system image segmentation neural network, training a segmentation network model, adjusting parameters, dividing an obtained image into a training image and a test image according to a ratio of 8:2, training the network by using the training image and a corresponding label file thereof in a model training stage, and obtaining trained weight parameters to segment the test image;
and training the deep learning segmentation network model by using a sample set similar to the root system to obtain corresponding training parameters.
6. The method for detecting the in-situ root system image of the plant based on the micro root canal as claimed in claim 5, wherein the network model is segmented, and the implementation mode is as follows:
building a U-net network combined with a transfer learning and attention mechanism:
setting a main stem extraction part in the U-net as a VGG16 network for extracting the characteristics of the image;
firstly, in the process of extracting a main trunk of an image, performing twice 64-channel convolution and Relu activation functions on a preprocessed image with the size of 512 × 3 by using VGG16, wherein 512 × 512 is the image scale, 3 is the number of channels of the image, a primary effective feature layer C1 is obtained, then performing 2 × 2 maximum pooling, wherein the maximum pooling does not change the number of channels of the image, and the maximum pooling compresses the height and width of the image to half to obtain 256 × 64 images;
performing two 128-channel convolution and Relu activation functions on the obtained 256 × 64 images to obtain an effective feature layer C2, and continuing to perform 2 × 2 maximal pooling, wherein the maximal pooling compresses the height and width of the images to half to obtain 128 × 128 images;
thirdly, performing 256-channel convolution and Relu activation functions on the obtained 128 × 128 images for three times to obtain an effective feature layer C3, and continuing to perform 2 × 2 maximum pooling, wherein the maximum pooling can compress the height and width of the images to half to obtain 64 × 256 images;
and fourthly, performing convolution of the obtained 64 x 256 images by three 512 channels and a Relu activation function to obtain an effective feature layer C4, and continuing to perform maximum pooling of 2 x 2, wherein the maximum pooling compresses the height and width of the images to half to obtain 32 x 512 images.
Fifthly, performing convolution with the step length of 1 x 1 and the channel of 512 on the image obtained in the step (iv) for three times to obtain a feature layer K of 32 x 512;
adding an attention mechanism into the U-net upsampling, overlapping the K and an effective feature layer C4 connected in a jumping mode with the upsampled K after the K passes through the attention mechanism, generating K1 after two 3 x 3 convolutions and Relu activation functions, overlapping the K1 and the C3 with the upsampled K1 after passing through the attention mechanism, generating K2 after the same two 3 x 3 convolutions and the Relu activation functions, repeating the steps twice, and obtaining a feature map with the same size as the input image through one 1 x 1 convolution;
migrating model parameters obtained by a sample set similar to a root system by using migration learning, inputting training data into a U-net network combined with the migration learning and attention mechanism for parameter training, and storing the parameters obtained by training; and finally, segmenting the test data set by using the trained weight to obtain a segmentation result graph.
7. The method for detecting the in-situ root system image of the plant based on the micro root canal as claimed in claim 1, further comprising a prediction network, wherein the implementation mode is as follows:
detecting detection data by using a model file obtained by a U-net network combining a training attention mechanism to realize prediction of root system segmentation;
calculating an AUC value through an ROC curve in each training period, and judging the quality of the model by using the AUC value, wherein the method is specifically realized as follows:
formula for false positive rate FPR:
equation of true positive ratio TPR:
in the formula (1), FPR is the horizontal axis of the ROC characteristic curve; FP is originally a non-root sample, but is identified by the network as a root sample; n is all non-root samples;
in the formula (2), TPR is the vertical axis of the ROC characteristic curve; TP is identifying the correct root sample; p is all root samples;
the larger the TPR is, the higher the proportion of the positive samples which are correctly identified is, and the smaller the FPR is, the smaller the proportion of the negative samples which are wrongly identified as the positive samples is;
the AUC is the area under the ROC curve, is between 0.1 and 1, and is used as the quality of a numerical evaluation model, and the larger the value is, the better the model is represented.
8. The apparatus for detecting the in-situ root system image of the plant based on the micro root canal is characterized by comprising a memory and a processor, wherein the memory stores a computer program, and the processor realizes the steps of the method for detecting the in-situ root system image of the plant based on the micro root canal according to any one of claims 1 to 7 when executing the computer program.
9. A computer-readable storage medium having stored thereon a computer program, characterized in that: the computer program when executed by a processor implements the method for in situ root system image detection of a micro root canal based plant according to any one of claims 1 to 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111210073.XA CN113935972A (en) | 2021-10-18 | 2021-10-18 | Method and device for detecting root system image of plant in situ based on micro root canal and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111210073.XA CN113935972A (en) | 2021-10-18 | 2021-10-18 | Method and device for detecting root system image of plant in situ based on micro root canal and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113935972A true CN113935972A (en) | 2022-01-14 |
Family
ID=79279855
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111210073.XA Pending CN113935972A (en) | 2021-10-18 | 2021-10-18 | Method and device for detecting root system image of plant in situ based on micro root canal and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113935972A (en) |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113421276A (en) * | 2021-07-02 | 2021-09-21 | 深圳大学 | Image processing method, device and storage medium |
-
2021
- 2021-10-18 CN CN202111210073.XA patent/CN113935972A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113421276A (en) * | 2021-07-02 | 2021-09-21 | 深圳大学 | Image processing method, device and storage medium |
Non-Patent Citations (5)
Title |
---|
ANINDYA APRILIYANTI PRAVITASARI,ET AL.: "UNet-VGG16 with transfer learning for MRI-based brain tumor segmentation", 《TELKOMNIKA TELECOMMUNICATION, COMPUTING, ELECTRONICS AND CONTROL 》, 30 June 2020 (2020-06-30) * |
JIA KANG,ET AL.: "Semantic segmentation model of cotton roots in-situ image based on attention mechanism", 《COMPUTERS AND ELECTRONICS IN AGRICULTURE》, 12 August 2021 (2021-08-12), pages 2 * |
OZAN OKTAY,ET AL.: "Attention U-Net: Learning Where to Look for the Pancreas", 《ARXIV》, 20 May 2018 (2018-05-20), pages 2 * |
任友群主编: "《音频音乐与计算机的交融 音频音乐技术》", 31 January 2020, 复旦大学出版社, pages: 102 - 104 * |
刘宏志编著: "《推荐系统》", 31 May 2020, 机械工业出版社, pages: 90 - 91 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Chen et al. | Detecting citrus in orchard environment by using improved YOLOv4 | |
CN114387520B (en) | Method and system for accurately detecting compact Li Zijing for robot picking | |
CN109902761B (en) | Fishing situation prediction method based on marine environment factor fusion and deep learning | |
CN110472597A (en) | Rock image rate of decay detection method and system based on deep learning | |
CN113744226A (en) | Intelligent agricultural pest identification and positioning method and system | |
CN111898677A (en) | Plankton automatic detection method based on deep learning | |
CN117576195A (en) | Plant leaf morphology recognition method | |
Lu et al. | Citrus green fruit detection via improved feature network extraction | |
Rappazzo et al. | EeLISA: Combating global warming through the rapid analysis of eelgrass wasting disease | |
Ärje et al. | Automatic flower detection and classification system using a light-weight convolutional neural network | |
Karakaya et al. | Discovery of agricultural diseases by deep learning and object detection. | |
CN113935972A (en) | Method and device for detecting root system image of plant in situ based on micro root canal and storage medium | |
KR20210028966A (en) | Method and apparatus for disease classification of plant leafs | |
CN116310541A (en) | Insect classification method and system based on convolutional network multidimensional learning | |
CN115063602A (en) | Crop pest and disease identification method based on improved YOLOX-S network | |
John et al. | Using photographs and deep neural networks to understand flowering phenology and diversity in mountain meadows | |
Perez et al. | Using spatiotemporal information in weather radar data to detect and track communal roosts | |
Stanski et al. | Flower detection using object analysis: new ways to quantify plant phenology in a warming tundra biome | |
Goyal et al. | Mining plants features for disease detection tensor flow: a boon to agriculture | |
CN117456287B (en) | Method for observing population number of wild animals by using remote sensing image | |
Li et al. | MCFF: Plant leaf detection based on multi-scale CNN feature fusion | |
CN109632799A (en) | The mobile detection stage division of rice leaf nitrogen content based on machine vision, system and computer readable storage medium | |
CN117765410B (en) | Remote sensing image double-branch feature fusion solid waste identification method and system and electronic equipment | |
Kurbanov et al. | Algorithm for recognizing and measuring parameters of biological objects in agriculture based on deep learning convolutional neural networks | |
Wu et al. | CCHA YOLO for mycelium clamp connection (CC) and hyphae Autolysis (HA) detection under microscopy imaging and web deployment |
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 |