CN115830474A - Method and system for identifying wild Tibetan medicine lamiophlomis rotata and distribution thereof and calculating yield thereof - Google Patents
Method and system for identifying wild Tibetan medicine lamiophlomis rotata and distribution thereof and calculating yield thereof Download PDFInfo
- Publication number
- CN115830474A CN115830474A CN202111264363.2A CN202111264363A CN115830474A CN 115830474 A CN115830474 A CN 115830474A CN 202111264363 A CN202111264363 A CN 202111264363A CN 115830474 A CN115830474 A CN 115830474A
- Authority
- CN
- China
- Prior art keywords
- unique
- distribution
- yield
- image
- area
- 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
Images
Abstract
The invention provides a method and a system for identifying wild Tibetan medicine lamiophlomis rotata and distribution thereof and calculating yield thereof, belonging to the field of traditional Chinese medicine resource investigation. The method comprises the following steps: (1) image acquisition; (2) processing and labeling pictures; (3) a training stage; and (4) identifying the unique flavor and the distribution thereof and calculating the yield thereof. The method for identifying the wild Tibetan medicine lamiophlomis rotata and the distribution thereof and calculating the yield of the wild Tibetan medicine lamiophlomis rotata improves the identification rate of small target plants, particularly the wild Tibetan medicine lamiophlomis rotata, correctly generates corresponding masks at the positioning positions, overcomes the problems of identifying the small target herbaceous plants by combining an image identification technology with the existing unmanned aerial vehicle, really applies the method to the small target herbaceous plants, particularly the wild Tibetan medicine lamiophlomis rotata, improves the identification rate of the wild Tibetan medicine lamiophlomis rotata, and can effectively calculate the yield of the distribution areas of the wild Tibetan medicine lamiophlomis rotata, and provides a reliable technology for traditional Chinese medicine resource investigation and yield monitoring.
Description
Technical Field
The invention belongs to the field of traditional Chinese medicine resource investigation, and particularly relates to a method and a system for identifying wild Tibetan medicine lamiophlomis rotata and distribution thereof and calculating yield of the wild Tibetan medicine lamiophlomis rotata.
Background
The investigation of the yield of wild medicinal plants is a key link and a difficult problem of sustainable utilization of traditional Chinese medicine resources. At present, the method mainly adopts sample prescription survey and combines the 3S technology to estimate the yield, and the method is widely applied to Chinese medicine resource survey. However, the traditional field sample biomass survey is time-consuming and labor-consuming and has limited survey area, and the distribution of wild medicinal plant resources is often uneven, so that the efficiency, the accuracy and the applicability of the method for surveying the reserve volume are greatly improved, and the method is particularly prominent in the survey of medicinal resources in severe regions such as plateau, desert and the like in China.
In recent years, image machine vision recognition is in blowout development with the progress of artificial intelligence technology, and the progress is rapid in face recognition, unmanned driving and medical image recognition. For example, machine vision recognition of CT images is realized based on an artificial intelligence algorithm, so that diagnosis efficiency is greatly improved when disease diagnosis is carried out, and huge development potential and advantages are embodied.
Meanwhile, the technology and application of Unmanned Aerial vehicles are also rapidly developed, and the Unmanned Aerial vehicles are collectively called Unmanned Aerial Vehicles (UAVs), which are emerging directions with rapid hardware development and application. Unmanned aerial vehicle surveying is rapidly instrumental, being considered as an "aerial sensor". Unmanned aerial vehicle remote sensing photography is unmanned aerial vehicle's important application, has advantages such as high mobility, high ageing, high resolution, cloud flight and low cost, uses extensively in fields such as agriculture, forestry, resource, ecology, environmental protection, can carry out regional image acquisition work fast to combine ground measured data, accomplish the species information monitoring in this region, and provide the precision verification for remote sensing on a large scale.
Therefore, the wild medicinal plants are expected to be identified by adopting the unmanned aerial vehicle technology and combining with image machine vision identification, and the distribution and the yield of the wild medicinal plants are analyzed and counted. Patent CN107578447B discloses a crop ridge position determining method and system based on unmanned aerial vehicle image planting. The method specifically comprises the following steps: obtaining crop area images shot by an unmanned aerial vehicle, and removing non-crop pattern spots such as weeds and the like according to the extracted pattern spot area and the pattern spot area characteristics through binaryzation; extracting crop characteristic points based on a crop imaging principle; establishing an h x d rectangular window parallel to the crop ridge direction, and screening crop characteristic points meeting requirements through a filtering and scanning strategy; and determining the positions of the crop ridges by adopting a least square method according to the screened crop characteristic points. It can utilize unmanned aerial vehicle image to detect out the crop ridge accurately and confirm the crop ridge position. The patent CN110378303A discloses that the area of a plant to be identified can be accurately found in a picture by adopting methods such as Mask-RCNN and the like; the patent CN109887020B discloses that a Mask-RCNN deep learning method can be adopted to train a stem and leaf area detection model, and a two-dimensional projection point image of a plant to be detected is detected by using the stem and leaf area detection model to obtain a stem and leaf area of the plant to be detected.
The wild Tibetan medicine lamiophlomis rotata is a Tibetan medicine with high medicinal value, can accurately identify the variety of the lamiophlomis rotata and determine the distribution of the lamiophlomis rotata to calculate the yield of the lamiophlomis rotata, and has important significance for further development of the lamiophlomis rotata. However, survey of plateau lamiophlomis rotata resource sample parties is difficult, cannot fully cover a lamiophlomis rotata growing area, and is time-consuming and labor-consuming. At present, few methods for detecting the lamiophlomis rotata of the wild Tibetan medicine are available, the identification accuracy is low, and the specialty is not strong. There is a need for a method and system for rapidly and accurately measuring the distribution of lamiophlomis rotata in wild Tibetan medicine. The problem is expected to be solved by adopting the unmanned aerial vehicle technology and combining image machine vision identification. However, in the prior art, an unmanned aerial vehicle system is combined with Mask R-CNN, so that the unmanned aerial vehicle system only has a good identification effect on a large target, and the identification rate of small target medicinal plants (with the size of 2cm-30 cm) is low. Meanwhile, if the patent method is adopted to identify the unique taste, the unique taste and other similar plants can be identified uniformly, and the time and the process for realizing the unique taste identification and the yield calculation of the system are too long, so that the identification rate is low and the time consumption is long.
Disclosure of Invention
In order to solve the problems, the invention provides a method and a system for identifying the lamiophlomis rotata of the wild Tibetan medicine and the distribution thereof and calculating the yield of the lamiophlomis rotata.
The technical scheme of the invention is as follows:
the invention provides a method for identifying wild Tibetan medicine lamiophlomis rotata and distribution thereof and calculating yield thereof, which comprises the following steps:
(1) Image acquisition: shooting an RGB image of the unique distribution area by adopting an unmanned aerial vehicle;
(2) Picture processing and labeling: cutting and marking the shot RGB image, converting a marking sample data set into a format of a binary mask label, dividing the format into a training set, a test set and a verification set, and using the training set, the test set and the verification set for model training;
(3) A training stage: carrying out model training by using a Python programming language under a Tensorflow deep learning framework until val _ loss converges to obtain a model for identifying the unique taste and the distribution of the unique taste and calculating the yield of the wild Tibetan medicine;
(4) Identifying the unique taste and its distribution and calculating its yield: and (4) processing the RGB images of the unique distribution area to generate an orthophoto map, sending the orthophoto map into the model in the step (3) for recognition, splicing the recognized images, and finally calculating the area and the yield of the distribution area.
Further, the air conditioner is provided with a fan,
in the step (1), the image acquisition comprises the steps of acquiring images of the lamiophlomis rotata in different growth stages, different weather conditions, different deteriorated grasslands and different flight heights;
and/or in the step (2), the clipping is performed by using an open-cv module in Python;
and/or in the step (2), marking by adopting a Labelme open source marking tool as marking software to produce a json file data set;
and/or in the step (2), the proportion of the training set, the test set and the verification set is 7;
and/or, in the step (4), the method for generating the orthophoto map by processing the RGB image of the unique distribution area comprises the steps of adding the RGB image into Agisoft Metascape Pro software, calibrating, aligning photos, selecting ultrahigh precision, inputting mark coordinates for coordinate position calibration, optimizing camera alignment, establishing dense point cloud, setting the highest quality, establishing DSM (digital surface model), and finally generating the orthophoto map;
and/or, in step (4), the orthophotographic map is cropped to 512pixel by 512pixel size;
and/or, in the step (4), the method for calculating the area and the yield of the distribution area is to divide the identified unique taste from the background by an ENVI5.3 software image processing technology to calculate the total leaf area of the unique taste in the distribution area, and calculate the yield of the distribution area by a linear relation between the fresh weight of the overground part and the leaf area.
Further, the air conditioner is provided with a fan,
in the step (1), the different growth stages comprise a lamiophlomis rotata seedling stage, a mature stage, a flowering stage and a fruiting stage;
and/or, in step (1), the different deteriorated grass comprises non-deteriorated grass, light deteriorated grass, medium deteriorated grass and heavy deteriorated grass;
and/or in the step (1), the different flying heights are 10m, 15m, 20m and 25m of the flying height of the unmanned aerial vehicle.
Further, the air conditioner is provided with a fan,
in the step (3), the model training comprises the following steps:
1) Scaling the image to the size of 512pixel by using a bilinear interpolation mode, and inputting the image into a ResNet network;
2) Constructing a plurality of candidate RoI areas for each element point of each layer of feature map in the image pyramid;
3) Respectively carrying out coordinate regression and foreground and background classification on each RoI region by using an RPN (resilient packet network);
4) Selecting a matched RoI area for each real area;
5) Sorting the areas judged as the foreground according to an intersection ratio IoU and an output value score of an RPN network, and selecting a plurality of RoI areas with the highest IoU and the highest score value as matching areas of real areas;
6) Finally, inputting the result into a segmentation network, a coordinate regression network and a classification network for training until val _ loss converges;
preferably, the first and second electrodes are formed of a metal,
in the step (3), the long size of the RoI region is formed by pairwise combination of the proportion of (0.5, 1, 2) and the length of (8, 16,32,64, 128), and one element point comprises 15 RoI regions.
The present invention also provides an apparatus for recognizing and calculating the distribution of lamiophlomis rotata of a wild Tibetan medicine, the apparatus comprising:
A. the unique distribution area image acquisition module is used for acquiring an RGB image of the unique distribution area shot by the unmanned aerial vehicle;
B. the image clipping module is used for clipping the acquired RGB image of the unique distribution area;
C. the data acquisition module is used for acquiring image data, converting an labeled sample data set into a format of a binary mask label after labeling, dividing the labeled sample data set into a training set, a test set and a verification set and using the training set, the test set and the verification set for model training;
D. the data learning module is used for carrying out model training on the collected unique image data, and the data learning module is carried out under a Tensorflow deep learning framework through a Python programming language until val _ loss converges to obtain a model for identifying the unique flavor of the wild Tibetan medicine and the distribution thereof and calculating the yield of the unique flavor;
E. a module to identify the unique and its distribution and calculate its yield: processing the RGB images of the unique distribution area to generate an orthophoto map, sending the orthophoto map into a model for identification, splicing the identified images, and finally calculating the area and the yield of the distribution area;
F. and the result display module is used for displaying the final result.
Further, the air conditioner is provided with a fan,
in the unique distribution area image acquisition module, the RGB images comprise RGB images of the unique with different growth stages, different weather conditions, different deteriorated grasslands and different flying heights;
and/or in the image cropping module, the cropping is performed by using an open-cv module in Python;
and/or in the data acquisition module, the label is marked by adopting a Labelme open source marking tool as marking software to produce a json file data set;
and/or in the data acquisition module, the proportion of the training set, the test set and the verification set is 7;
and/or, in the module for identifying the unique taste and the distribution thereof and calculating the yield thereof, the method for generating the orthophotograph by processing the RGB image of the unique taste distribution area comprises the steps of adding the RGB image into Agisoft Metashape Pro software, calibrating, aligning photos, selecting ultrahigh precision, inputting a mark coordinate for coordinate position calibration, then optimizing camera alignment, establishing dense point cloud, setting the quality to be the highest, establishing DSM, and finally generating the orthophotograph;
and/or, in the module that identifies the unique and its distribution and calculates its yield, the orthophotomap is cropped to 512pixel × 512pixel size;
and/or in the module for identifying the unique taste and the distribution thereof and calculating the yield of the unique taste, the method for calculating the area and the yield of the distribution area comprises the steps of segmenting the identified unique taste from the background by an ENVI5.3 software image processing technology to calculate the total leaf area of the unique taste in the distribution area, and calculating the yield of the distribution area by the linear relation between the fresh weight of the overground part and the leaf area.
Further, the air conditioner is provided with a fan,
in the image acquisition module of the unique flavor distribution area, the different growth stages comprise a unique flavor seedling stage, a mature stage, a flowering stage and a fruiting stage;
and/or in the unique distribution area image acquisition module, wherein the different deteriorated grasslands comprise non-deteriorated grasslands, slightly deteriorated grasslands, moderately deteriorated grasslands and severely deteriorated grasslands;
and/or in the unique distribution area image acquisition module, the different flying heights are 10m, 15m, 20m and 25m of the flying height of the unmanned aerial vehicle.
Further, the air conditioner is provided with a fan,
a data learning module, the model training comprising the steps of:
1) Scaling the image to the size of 512pixel by using a bilinear interpolation mode, and inputting the image into a ResNet network;
2) Constructing a plurality of candidate RoI areas for each element point of each layer of feature map in the image pyramid;
3) Respectively carrying out coordinate regression and foreground and background classification on each RoI region by using an RPN (resilient packet network);
4) Selecting a matched RoI area for each real area;
5) Sorting the areas judged as the foreground according to an intersection ratio and an output value score of an RPN network, and selecting a plurality of RoI areas with the highest IoU and the highest score value as matching areas of real areas;
6) Finally, inputting the result into a segmentation network, a coordinate regression network and a classification network for training until val _ loss converges;
preferably, the first and second electrodes are formed of a metal,
in the data learning module, the long size of the RoI region is formed by pairwise combination of the proportion of (0.5, 1, 2) and the length of (8, 16,32,64, 128), and one element point comprises 15 RoI regions.
The invention also provides a computer-readable storage medium having stored thereon a computer program for causing a computer to execute a method as described above.
The invention also provides a system for identifying the wild Tibetan medicine lamiophlomis rotata and the distribution thereof and calculating the yield thereof, which comprises the following equipment connected by a data line and/or a data interface:
a unique distribution area RGB image capture and/or input device and the aforementioned devices.
Firstly, an unmanned aerial vehicle is used for obtaining RGB images of a unique distribution area on plateau and grassland, cutting is carried out on the original scale, the distribution position of the cut unique is marked by using labelme, and the distribution position is divided into a training set, a testing set and a verification set; selecting a rapid residual error network ResNet-101 in Mask-RCNN for training until val _ loss converges; splicing unmanned aerial vehicle images in a unique distribution area by using Agisoft Metascape Pro, sending spliced pictures into a model for unique identification, calculating the proportion of the mask area of the unique to the area of the whole distribution area, finally performing real area conversion, and obtaining the unique yield of the final distribution area through the processes of cutting-identification-splicing-yield calculation.
Because the small-target herbaceous plants are mostly distributed in a messy way, the background is complex, and the small-target herbaceous plants are difficult to distinguish accurately, an unmanned aerial vehicle system and a Mask R-CNN in the prior art are combined at present, the large target herbaceous plants can be recognized well, and the small-target herbaceous plants are low in distinguishing rate. The method for identifying the wild Tibetan medicine lamiophlomis rotata and the distribution thereof and calculating the yield of the wild Tibetan medicine lamiophlomis rotata improves the identification rate of small target plants, particularly the wild Tibetan medicine lamiophlomis rotata, correctly generates corresponding masks at the positioning positions, overcomes the problems of the identification of the small target herbaceous plants by the existing unmanned aerial vehicle combined with the image identification technology, really applies the wild Tibetan medicine lamiophlomis rotata to the small target herbaceous plants, particularly to the identification of the wild Tibetan medicine lamiophlomis rotata, improves the identification rate of the wild Tibetan medicine lamiophlomis rotata, and can effectively calculate the yield of the distribution area of the wild Tibetan medicine lamiophlomis rotata.
Obviously, many modifications, substitutions, and variations are possible in light of the above teachings of the invention, without departing from the basic technical spirit of the invention, as defined by the following claims.
The present invention will be described in further detail with reference to the following examples. This should not be understood as limiting the scope of the above-described subject matter of the present invention to the following examples. All the technologies realized based on the above contents of the present invention belong to the scope of the present invention.
Drawings
FIG. 1 shows that Labelme software is used to manually label the plants with the unique flavor, and the growing range of each plant in the image is circled, wherein A 1 And A 2 As an original image, B 1 And B 2 To mark the location of a unique taste, C 1 And C 2 And displaying the marked result.
FIG. 2 is a flow chart of training recognition: firstly, mapping the extracted unique image features into a support network; then, a region suggestion network (RPN) that generates a region of interest (ROI) from the feature map formed by the skeleton; extracting corresponding target features from the RPN by a Full Convolution Network (FCN), and then classifying and segmenting the target; the output of this stage is the generation of the classification score, the segmentation mask and the bounding box.
FIG. 3 shows the results of identifying the Lamiophlomis rotata of the wild Tibetan medicine and the distribution thereof: a is the recognition result of the moderate deteriorated grassland, and B is the recognition result of the severe deteriorated grassland; the lower left corner in the figure is the recognition result of the whole ortho-image, and the upper right corner is the recognition result of partial interception.
FIG. 4 is a distribution diagram of identified unique species processed by the ENVI5.3 software image processing technique.
Detailed Description
Example 1 method for identifying and calculating the distribution and yield of Lamiophlomis rotata of wild Tibetan medicine
Step 3, marking pictures: adopting a Labelme open source labeling tool as labeling software, wherein labeling is to manually circle out the unique flavor, generating a json file data set after labeling, converting a labeling sample data set into a format of a binary mask label, and performing the following steps of: 2: the system 1 is divided into a training set, a testing set and a verification set and is used for model training and adjustment.
Step 4, training: performed under the Tensorflow deep learning framework via the Python programming language. Firstly, images are scaled to the size of 512pixel by using a bilinear interpolation mode, then the images are input into a ResNet network, a plurality of candidate RoI areas are constructed for each element point of each layer of feature map in an image pyramid, the long dimension of the RoI area is formed by pairwise combination of the proportion of (0.5, 1, 2) and the length of (8, 16,32,64, 128), and the total number of 15 RoI areas is one element point. Coordinate regression and foreground and background classification are performed on each RoI area with the RPN network, respectively. Selecting a matched RoI area for each real area, sorting the areas judged as the foreground according to an intersection of Union (IoU) and an output value score of an RPN network, selecting a plurality of RoI areas with the highest IoU and the highest score value as the matching areas of the real areas, and finally inputting the matching areas into a segmentation network, a coordinate regression network and a classification network for training. The learning rate and epoch of the model are adjusted by val _ loss until the convergence of val _ loss, which is the optimal weight value of the model.
Step 5, orthographic projection image mapping: the method comprises the steps of arranging shot aerial photos of the unmanned aerial vehicle in the wild unique distribution area, adding the aerial photos into Agisoft Metascape Pro software, checking camera calibration, aligning photos, selecting ultrahigh precision, inputting mark coordinates to calibrate coordinate positions, optimizing camera alignment, establishing dense point cloud, setting up DSM (digital surface model), and finally generating an orthophoto map to be exported with the highest quality.
Step 6, model identification: and (4) sending the unique orthophoto map into the model obtained in the step (4), wherein the pretreatment mainly comprises the steps of cutting the orthophoto map into 512pixel by 512pixel before the model identification, then sending each image into the model for identification, and splicing the identified images to obtain the unique orthophoto map after identification.
And 7, calculating the area yield: and (3) segmenting the identified unique taste from the background by an ENVI5.3 software image processing technology to calculate the total leaf area of the unique taste in the distribution area, and calculating the yield of the distribution area by the linear relation between the fresh weight of the overground part and the leaf area.
Example 2 identification of wild Tibetan drug Lamiophlomis rotata and its distribution and calculation of its yield by the method of the present invention
By adopting the method of the embodiment 1, the distribution of the unique flavor of the wild Tibetan medicine at the watershed of the Abaga county is identified, and the yield of the wild Tibetan medicine is calculated. The results obtained were: the number of identified lamiophlomis rotata plants is 27257, and the distribution leaf area is 183.0m 2 The calculated yield was 98.0kg.
And (5) similarly identifying the distribution of the wild Tibetan medicine unique flavor in the place by adopting a sample prescription investigation method, and calculating the yield of the wild Tibetan medicine unique flavor. The results obtained were: the number of lamiophlomis rotata strains is 21222.65, and the calculated yield is 46.65587kg.
The actual number of wild Tibetan medicine lamiophlomis rotata strains at watershed of Abam county is 25137, and the yield is 85kg. Compared with the two methods, the unmanned aerial vehicle identification result is more accurate than the identification result of a sample survey method and is close to a practical result.
In conclusion, because small-target herbaceous plants are mostly distributed in a messy manner, the background is complex, and accurate identification is difficult, an unmanned aerial vehicle system and a Mask R-CNN in the prior art are combined at present, so that a good identification effect can be achieved only for large targets, and the identification rate for small-target medicinal plants is low. The method for identifying the wild Tibetan medicine lamiophlomis rotata and the distribution thereof and calculating the yield of the wild Tibetan medicine lamiophlomis rotata improves the identification rate of small target plants, particularly the wild Tibetan medicine lamiophlomis rotata, correctly generates corresponding masks at the positioning positions, overcomes the problems of the identification of the small target herbaceous plants by the existing unmanned aerial vehicle combined with the image identification technology, really applies the wild Tibetan medicine lamiophlomis rotata to the small target herbaceous plants, particularly to the identification of the wild Tibetan medicine lamiophlomis rotata, improves the identification rate of the wild Tibetan medicine lamiophlomis rotata, and can effectively calculate the yield of the distribution area of the wild Tibetan medicine lamiophlomis rotata.
Claims (10)
1. A method for identifying wild Tibetan medicine lamiophlomis rotata and distribution thereof and calculating yield thereof is characterized in that: it comprises the following steps:
(1) Image acquisition: shooting an RGB image of the unique distribution area by adopting an unmanned aerial vehicle;
(2) Picture processing and labeling: cutting and marking the shot RGB image, converting a marking sample data set into a format of a binary mask label, dividing the format into a training set, a test set and a verification set, and using the training set, the test set and the verification set for model training;
(3) A training stage: carrying out model training by using a Python programming language under a Tensorflow deep learning framework until val _ loss converges to obtain a model for identifying the unique taste and the distribution of the unique taste and calculating the yield of the wild Tibetan medicine;
(4) Identifying the unique taste and its distribution and calculating its yield: and (4) processing the RGB images of the unique distribution area to generate an orthophoto map, sending the orthophoto map into the model in the step (3) for identification, splicing the identified images, and finally calculating the area and the yield of the distribution area.
2. The method of claim 1, wherein:
in the step (1), the image acquisition comprises the steps of acquiring images of the lamiophlomis rotata in different growth stages, different weather conditions, different deteriorated grasslands and different flight heights;
and/or in the step (2), the clipping is performed by using an open-cv module in Python;
and/or in the step (2), marking by adopting a Labelme open source marking tool as marking software to produce a json file data set;
and/or in the step (2), the proportion of the training set, the test set and the verification set is 7;
and/or, in the step (4), the method for generating the orthophoto map by processing the RGB image of the unique distribution area comprises the steps of adding the RGB image into Agisoft Metascape Pro software, calibrating, aligning photos, selecting ultrahigh precision, inputting mark coordinates for coordinate position calibration, optimizing camera alignment, establishing dense point cloud, setting the highest quality, establishing DSM (digital surface model), and finally generating the orthophoto map;
and/or, in step (4), the orthophotographic map is cropped to 512pixel by 512pixel size;
and/or, in the step (4), the method for calculating the area and the yield of the distribution area is to divide the identified unique taste from the background by an ENVI5.3 software image processing technology to calculate the total leaf area of the unique taste in the distribution area, and calculate the yield of the distribution area by a linear relation between the fresh weight of the overground part and the leaf area.
3. The method of claim 2, wherein:
in the step (1), the different growth stages comprise a lamiophlomis rotata seedling stage, a mature stage, a flowering stage and a fruiting stage;
and/or, in step (1), the different deteriorated grass comprises non-deteriorated grass, slightly deteriorated grass, moderately deteriorated grass and heavily deteriorated grass;
and/or in the step (1), the different flying heights are 10m, 15m, 20m and 25m of the flying height of the unmanned aerial vehicle.
4. The method of claim 1, wherein:
in the step (3), the model training comprises the following steps:
1) Scaling the image to the size of 512pixel by using a bilinear interpolation mode, and inputting the image into a ResNet network;
2) Constructing a plurality of candidate RoI areas for each element point of each layer of feature map in the image pyramid;
3) Performing coordinate regression and foreground and background classification on each RoI region by using an RPN (resilient packet network) network respectively;
4) Selecting a matched RoI area for each real area;
5) Sorting the areas judged as the foreground according to an intersection ratio IoU and an output value score of an RPN network, and selecting a plurality of RoI areas with the highest IoU and the highest score value as matching areas of real areas;
6) Finally, inputting the result into a segmentation network, a coordinate regression network and a classification network for training until val _ loss converges;
preferably, the first and second electrodes are formed of a metal,
in the step (3), the long size of the RoI region is formed by pairwise combination of the proportion of (0.5, 1, 2) and the length of (8, 16,32,64, 128), and one element point comprises 15 RoI regions.
5. An apparatus for identifying and calculating the distribution and yield of a wild Tibetan drug Lamiophlomis rotata, comprising: the apparatus comprises:
A. the unique distribution area image acquisition module is used for acquiring an RGB image of the unique distribution area shot by the unmanned aerial vehicle;
B. the image clipping module is used for clipping the acquired RGB image of the unique distribution area;
C. the data acquisition module is used for acquiring image data, converting an labeled sample data set into a format of a binary mask label after labeling, dividing the labeled sample data set into a training set, a test set and a verification set and using the training set, the test set and the verification set for model training;
D. the data learning module is used for carrying out model training on the collected unique image data, and the data learning module is carried out under a Tensorflow deep learning framework through a Python programming language until val _ loss converges to obtain a model for identifying the unique flavor of the wild Tibetan medicine and the distribution thereof and calculating the yield of the unique flavor;
E. a module to identify the unique and its distribution and calculate its yield: processing the RGB images of the unique distribution area to generate an orthophoto map, sending the orthophoto map into a model for identification, splicing the identified images, and finally calculating the area and the yield of the distribution area;
F. and the result display module is used for displaying the final result.
6. The apparatus of claim 5, wherein:
in the unique distribution area image acquisition module, the RGB images comprise RGB images of the unique with different growth stages, different weather conditions, different deteriorated grasslands and different flying heights;
and/or in the image cropping module, the cropping is performed by using an open-cv module in Python;
and/or in the data acquisition module, the label is marked by adopting a Labelme open source marking tool as marking software to produce a json file data set;
and/or in the data acquisition module, the proportion of the training set, the test set and the verification set is 7;
and/or, in the module for identifying the unique taste and the distribution thereof and calculating the yield thereof, the method for generating the orthophotograph by processing the RGB image of the unique taste distribution area comprises the steps of adding the RGB image into Agisoft Metascape Pro software, calibrating, aligning photos, selecting ultrahigh precision, inputting mark coordinates for coordinate position calibration, then optimizing camera alignment, establishing dense point cloud, setting the quality to be highest, establishing DSM (digital surface model), and finally generating the orthophotograph;
and/or, in the module that identifies the unique and its distribution and calculates its yield, the orthophotomap is cropped to 512pixel × 512pixel size;
and/or in the module for identifying the unique taste and the distribution thereof and calculating the yield of the unique taste, the method for calculating the area and the yield of the distribution area comprises the steps of segmenting the identified unique taste from the background by an ENVI5.3 software image processing technology to calculate the total leaf area of the unique taste in the distribution area, and calculating the yield of the distribution area by the linear relation between the fresh weight of the overground part and the leaf area.
7. The apparatus of claim 6, wherein:
in the image acquisition module of the unique flavor distribution area, the different growth stages comprise a unique flavor seedling stage, a mature stage, a flowering stage and a fruiting stage;
and/or in the unique distribution area image acquisition module, wherein the different deteriorated grasslands comprise non-deteriorated grasslands, slightly deteriorated grasslands, moderately deteriorated grasslands and severely deteriorated grasslands;
and/or in the unique distribution area image acquisition module, the different flying heights are 10m, 15m, 20m and 25m of the flying height of the unmanned aerial vehicle.
8. The apparatus of claim 5, wherein:
a data learning module, the model training comprising the steps of:
1) Zooming the image to the size of 512pixel × 512pixel by using a bilinear interpolation mode, and inputting the image into a ResNet network;
2) Constructing a plurality of candidate RoI areas for each element point of each layer of feature map in the image pyramid;
3) Respectively carrying out coordinate regression and foreground and background classification on each RoI region by using an RPN (resilient packet network);
4) Selecting a matched RoI area for each real area;
5) Sorting the regions judged as the foreground according to the cross-over ratio and an output value score of an RPN network, and selecting a plurality of RoI regions with the highest IoU and the highest score value as matching regions of real regions;
6) Finally, inputting the result into a segmentation network, a coordinate regression network and a classification network for training until val _ loss converges;
preferably, the first and second electrodes are formed of a metal,
in the data learning module, the long size of the RoI region is formed by pairwise combination of the proportion of (0.5, 1, 2) and the length of (8, 16,32,64, 128), and one element point comprises 15 RoI regions.
9. A computer-readable storage medium characterized by: stored thereon a computer program for causing a computer to perform the method as claimed in any one of claims 1-4.
10. A system for identifying wild Tibetan medicine lamiophlomis rotata and distribution thereof and calculating yield thereof is characterized in that: the device comprises the following devices connected through a data line and/or a data interface:
a unique distribution area RGB image acquisition and/or input device and a device as claimed in any one of claims 5 to 8.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111082892 | 2021-09-15 | ||
| CN2021110828920 | 2021-09-15 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115830474A true CN115830474A (en) | 2023-03-21 |
Family
ID=85516032
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111264363.2A Pending CN115830474A (en) | 2021-09-15 | 2021-10-28 | Method and system for identifying wild Tibetan medicine lamiophlomis rotata and distribution thereof and calculating yield thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115830474A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116757507A (en) * | 2023-08-14 | 2023-09-15 | 武汉理工大学 | Crop grouting process prediction method, system, electronic equipment and storage medium |
-
2021
- 2021-10-28 CN CN202111264363.2A patent/CN115830474A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116757507A (en) * | 2023-08-14 | 2023-09-15 | 武汉理工大学 | Crop grouting process prediction method, system, electronic equipment and storage medium |
| CN116757507B (en) * | 2023-08-14 | 2023-11-10 | 武汉理工大学 | Crop grouting process prediction method, system, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104881865B (en) | Forest pest and disease monitoring method for early warning and its system based on unmanned plane graphical analysis | |
| CN110378909B (en) | Single wood segmentation method for laser point cloud based on Faster R-CNN | |
| CN106951836B (en) | crop coverage extraction method based on prior threshold optimization convolutional neural network | |
| CN109146889A (en) | A kind of field boundary extracting method based on high-resolution remote sensing image | |
| CN110070571B (en) | Phyllostachys pubescens morphological parameter detection method based on depth camera | |
| Baofeng et al. | Digital surface model applied to unmanned aerial vehicle based photogrammetry to assess potential biotic or abiotic effects on grapevine canopies | |
| CN111666855B (en) | Animal three-dimensional parameter extraction method and system based on unmanned aerial vehicle and electronic equipment | |
| CN108648169A (en) | The method and device of high voltage power transmission tower defects of insulator automatic identification | |
| CN109032174B (en) | Unmanned aerial vehicle operation route planning method and operation execution method | |
| CN106403904A (en) | Landscape-scale vegetation coverage calculation method and system based on unmanned aerial vehicle | |
| CN108711172B (en) | Unmanned aerial vehicle identification and positioning method based on fine-grained classification | |
| CN109813286A (en) | A kind of lodging disaster remote sensing damage identification method based on unmanned plane | |
| CN112285733B (en) | Urban and rural planning verification mapping data processing method | |
| CN112560623B (en) | Unmanned aerial vehicle-based rapid mangrove plant species identification method | |
| CN115331100A (en) | Spatial distribution monitoring method and system for cultivated land planting attributes | |
| Paturkar et al. | Non-destructive and cost-effective 3D plant growth monitoring system in outdoor conditions | |
| CN115115954A (en) | Intelligent identification method for pine nematode disease area color-changing standing trees based on unmanned aerial vehicle remote sensing | |
| Ming et al. | Cropland extraction based on OBIA and adaptive scale pre-estimation | |
| CN113033386B (en) | High-resolution remote sensing image-based transmission line channel hidden danger identification method and system | |
| CN115830474A (en) | Method and system for identifying wild Tibetan medicine lamiophlomis rotata and distribution thereof and calculating yield thereof | |
| CN111738159A (en) | Cotton terminal bud positioning method based on vector calibration | |
| CN110084120B (en) | Vegetation investigation method, device, system, equipment and medium in island environment | |
| CN116739739A (en) | Loan amount evaluation method and device, electronic equipment and storage medium | |
| Saeed et al. | Cotton plant part 3D segmentation and architectural trait extraction using point voxel convolutional neural networks | |
| CN116030324A (en) | Target detection method based on fusion of spectral features and spatial features |
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 |