CN114916336B - Chemical topping method based on cotton top leaf maturity stage classification and identification - Google Patents
Chemical topping method based on cotton top leaf maturity stage classification and identification Download PDFInfo
- Publication number
- CN114916336B CN114916336B CN202210486529.3A CN202210486529A CN114916336B CN 114916336 B CN114916336 B CN 114916336B CN 202210486529 A CN202210486529 A CN 202210486529A CN 114916336 B CN114916336 B CN 114916336B
- Authority
- CN
- China
- Prior art keywords
- topping
- cotton
- top leaf
- module
- cotton top
- 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.)
- Active
Links
- 235000019005 Digitaria californica Nutrition 0.000 title claims abstract description 95
- 241001115843 Digitaria californica Species 0.000 title claims abstract description 95
- 238000000034 method Methods 0.000 title claims abstract description 54
- 239000000126 substance Substances 0.000 title claims abstract description 22
- 229920000742 Cotton Polymers 0.000 claims abstract description 86
- 238000003062 neural network model Methods 0.000 claims abstract description 28
- 239000003795 chemical substances by application Substances 0.000 claims abstract description 15
- 230000008569 process Effects 0.000 claims abstract description 12
- 239000007921 spray Substances 0.000 claims abstract description 8
- 238000011176 pooling Methods 0.000 claims description 22
- 238000004220 aggregation Methods 0.000 claims description 14
- 230000002776 aggregation Effects 0.000 claims description 14
- 238000012216 screening Methods 0.000 claims description 14
- 238000000605 extraction Methods 0.000 claims description 13
- 238000012549 training Methods 0.000 claims description 10
- 238000005286 illumination Methods 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 7
- 238000009826 distribution Methods 0.000 claims description 5
- 230000000694 effects Effects 0.000 claims description 5
- 230000004927 fusion Effects 0.000 claims description 5
- 238000012360 testing method Methods 0.000 claims description 5
- 230000008859 change Effects 0.000 claims description 4
- 238000012795 verification Methods 0.000 claims description 4
- 238000005507 spraying Methods 0.000 claims description 3
- 239000002689 soil Substances 0.000 abstract description 5
- 241000219146 Gossypium Species 0.000 description 63
- 238000001514 detection method Methods 0.000 description 11
- 240000002024 Gossypium herbaceum Species 0.000 description 6
- 235000004341 Gossypium herbaceum Nutrition 0.000 description 6
- 241000196324 Embryophyta Species 0.000 description 2
- 230000015572 biosynthetic process Effects 0.000 description 2
- 239000013043 chemical agent Substances 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000003814 drug Substances 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 239000007788 liquid Substances 0.000 description 2
- 235000015097 nutrients Nutrition 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 241000238631 Hexapoda Species 0.000 description 1
- 241000607479 Yersinia pestis Species 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 239000003337 fertilizer Substances 0.000 description 1
- 230000003779 hair growth Effects 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000010297 mechanical methods and process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000000056 organ Anatomy 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000001850 reproductive effect Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- A—HUMAN NECESSITIES
- A01—AGRICULTURE; FORESTRY; ANIMAL HUSBANDRY; HUNTING; TRAPPING; FISHING
- A01G—HORTICULTURE; CULTIVATION OF VEGETABLES, FLOWERS, RICE, FRUIT, VINES, HOPS OR SEAWEED; FORESTRY; WATERING
- A01G7/00—Botany in general
- A01G7/06—Treatment of growing trees or plants, e.g. for preventing decay of wood, for tingeing flowers or wood, for prolonging the life of plants
-
- A—HUMAN NECESSITIES
- A01—AGRICULTURE; FORESTRY; ANIMAL HUSBANDRY; HUNTING; TRAPPING; FISHING
- A01G—HORTICULTURE; CULTIVATION OF VEGETABLES, FLOWERS, RICE, FRUIT, VINES, HOPS OR SEAWEED; FORESTRY; WATERING
- A01G3/00—Cutting implements specially adapted for horticultural purposes; Delimbing standing trees
- A01G3/08—Other tools for pruning, branching or delimbing standing trees
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Life Sciences & Earth Sciences (AREA)
- Biodiversity & Conservation Biology (AREA)
- Ecology (AREA)
- Forests & Forestry (AREA)
- Environmental Sciences (AREA)
- Engineering & Computer Science (AREA)
- Wood Science & Technology (AREA)
- Botany (AREA)
- Image Analysis (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention discloses a chemical topping method based on cotton topping mature stage classification recognition, which is characterized in that the chemical topping method is improved on the basis of the existing YOLOv4 neural network model, cotton topping machine is trained to obtain cotton topping classification recognition models based on different mature stages, cotton topping images are acquired through a camera module in the process of topping cotton topping by the cotton topping machine, and then the acquired data images are input into the cotton topping classification recognition models based on different mature stages to analyze and process the cotton topping images, and classification recognition results are output; the cotton topping machine can accurately topping cotton, so that topping agents can be saved, topping cost is reduced, and topping efficiency is improved. In addition, the cotton topping machine can accurately spray the topping agent onto cotton top leaves in the mature period, so that the topping agent can be prevented from polluting soil, and green operation is realized.
Description
Technical Field
The invention relates to the field of agriculture, in particular to a chemical topping method based on classification and identification of the maturity of cotton top leaves.
Background
Cotton has an infinite growth habit, so cotton needs to be topped, eliminating the top growth advantage of cotton. Cotton topping is one of the important links of the whole growth cycle of cotton and plays an important role in the yield of cotton. The cotton topping can adjust the transport direction of nutrient substances in plants, so that more nutrients are supplied to reproductive organs, the redundant loss of water and fertilizer is reduced, and early boll formation and multi-boll formation of cotton plants are promoted. The cotton topping is performed by picking off a center of the cotton.
Currently, in cotton topping operation by using a mechanical method, operators often do not stand at the angle of agricultural integration to perform cotton topping operation. For example, when an operator uses mechanical equipment to perform topping of cotton, the growth form of cotton plants in an actual environment is often not considered, and the phenomena of low accuracy, cutter leakage, over-cutting and the like exist in mechanized topping, so that the yield of cotton is seriously affected. And different cotton planting areas in China such as Xinjiang, shandong and the like still adopt manual topping and chemical topping. The biggest defects of manual topping are high labor intensity, long time and high labor cost; and the lack of labor force and low productivity ultimately affect cotton yield. Secondly, operators are easy to cause bud drop and spread of diseases and insect pests in the topping process. The chemical topping is to spray the topping agent onto the plant, and the topping agent has strong inhibition effect and can inhibit the hair growth of the whole cotton plant if being used inappropriately. Before using the topping agent, the topping agent is not sprayed at fixed points and quantitatively, so that the chemical agent can harm the soil, and good soil is a precondition for cotton plant production and growth, and the chemical agent is inappropriately used, so that the soil is easy to pollute and affect the next crop. Therefore, in the cotton topping process, the top leaf classification and identification system is required to be cooperated and corresponding to the cotton topping operation, so that the precise, medicine-saving and green operation on the basis of agricultural machinery and agriculture fusion is realized.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a chemical topping method based on the classification and identification of the mature period of cotton topping, which can cooperatively correspond a classification and identification system of the mature period of cotton topping with the existing topping mode so as to realize the precise identification of terminal buds and the precise, pesticide-saving and green operation based on agricultural integration in the mechanical or chemical topping process.
The technical scheme for solving the technical problems is as follows:
a chemical topping method based on classification and identification of top cotton leaf maturity comprises the following steps:
s1, acquiring cotton top leaf images of different maturity stages according to actual growth conditions of cotton on an operation land, wherein the steps include shooting cotton top leaf images of different time periods, shooting cotton top leaf images of different illumination angles and shooting cotton top leaf images containing different top leaf types;
s2, expanding the shot cotton top leaf image, dividing the expanded cotton top leaf image into a training set, a verification set and a test set according to the ratio of 8:1:1, manually marking by using a LabelImg tool, generating an xml type file, and storing the generated xml type file according to the format of the PASCALVOC data set;
s3, training cotton top leaf classification and identification models based on different maturity stages, wherein the cotton top leaf classification and identification models based on different maturity stages adopt a YOLOv4 neural network model, and the following improvements are made on the basis of the YOLOv4 neural network model:
s3.1, performing feature extraction on cotton top leaf targets in different maturity stages by taking CSPDarknet53 as a main feature extraction network, and performing five-time convolution downsampling by a residual error module to realize feature extraction and screening of input information, and outputting three different scale features to the next stage;
s3.2, using Spatial Pyramid Pooling (space pyramid pooling) module and Path Aggregation Network (path aggregation network) feature pyramid network in the neck part, and performing maximum pooling treatment on the last feature layer by using pooling layers of four different scales of 1×1,3×3,5×5 and 13×13 of SpatialPyramid Pooling (space pyramid pooling) modules; repeated up-sampling and down-sampling operations are carried out through a Path Aggregation Network (path aggregation network) module so as to realize fusion of the feature map information;
s3.3, introducing a coupled Head module into the Head module, and processing the final characteristic information again to obtain prediction frame information and confidence;
s3.4, introducing an Anchor Free method into the YOLOv4 neural network model, dividing the predicted frame information into 3 prior frames with different scales by using a label distribution method, and continuously screening out a positive sample prior frame closest to a real frame for final prediction; in the screening process, extracting the information of the positive sample prior frame, dynamically distributing the number of the positive samples by using a SimOTA method, and screening out the final positive sample prior frame by calculating the coincidence degree of the real frame and the positive sample prior frame and the type prediction accuracy data and outputting the final positive sample prior frame as a prediction result;
s4, in the process of topping cotton by the cotton topping machine, acquiring cotton top leaf images by a camera module, inputting the acquired data images into a main control module of a cotton top leaf classification and identification topping system in different maturity stages, analyzing and processing the cotton top leaf images by the main control module based on the cotton top leaf classification and identification models in different maturity stages, and outputting classification and identification results, wherein the main control module controls the cotton topping machine to accurately topping cotton.
Preferably, in step S2, the data enhancement method is used to perform data expansion in 5 modes of color enhancement, brightness enhancement, rotation, random saturation contrast sharpness and inversion on the original image, wherein the color enhancement, brightness enhancement and random saturation contrast sharpness are used to simulate different effects of the photographed cotton top leaf image caused by the field illumination intensity and angle change, and the inversion and rotation are used to simulate different viewing angles photographed by the camera.
Preferably, in step S4, the trained cotton top leaf classification recognition models based on different maturity stages are deployed by adopting an online deployment mode, so as to realize online real-time recognition of the cotton top leaf.
Preferably, in step S4, the cotton topping machine includes a high-clearance vehicle, a camera module disposed on the high-clearance vehicle, a topping module, and a main control module, where the topping module implements topping of cotton top leaves by spraying topping agent; the camera module is used for collecting cotton top leaf images and transmitting the collected top leaf images to the main control module, the main control module is provided with the cotton top leaf classification and identification models based on different maturity stages, the collected cotton top leaf images are analyzed and processed based on the cotton top leaf classification and identification models of different maturity stages to obtain the position information of the cotton top leaf needing topping, and then the topping module is controlled to spray topping agent to the cotton top leaf.
Preferably, the camera module and the topping module are arranged in a horizontal straight line, and the distance between the two groups is 0.05m, wherein the camera module is positioned in front of the topping module along the movement direction of the high-clearance vehicle.
Preferably, the movement speed of the high-clearance vehicle is 0.1m/s.
Preferably, in step (1), the captured top leaf image of the cotton includes mature top leaf alone, immature top leaf, combination of mature top leaf with mature top leaf, combination of immature top leaf with immature top leaf.
Preferably, the main control module is a single chip microcomputer.
Preferably, the camera module is a sony HDR-CX900E full high definition video camera.
Compared with the prior art, the invention has the following beneficial effects:
1. the topping method based on the classifying and identifying of the cotton top leaves in the mature period achieves the aim of topping based on classifying and identifying of the cotton top leaves in different mature periods from the technical point of view, provides a basis for intelligent cotton accurate topping, effectively improves cotton topping efficiency, meets the practical requirements that a planter does not want to waste liquid medicine and pollute soil because the cotton topping liquid medicine is excessively sprayed on other parts except top buds, and realizes accurate, medicine-saving and green operation.
2. The topping method based on the classifying and identifying of the mature period of the cotton top leaves reduces the labor intensity of workers, improves the working efficiency, reduces the topping cost of the cotton and effectively improves the yield of the cotton.
3. According to the topping method based on the classifying and identifying of the cotton top leaf maturity, in the aspect of improving the detection effectiveness of the YOLOv4 neural network model, the YOLOv4 neural network module is utilized to carry out a comparison test with the improved neural network model, and the result shows that the improved neural network model has good and more balanced performance in the aspect of detecting the top buds in different maturity, and has better detection performance and better performance.
Drawings
FIG. 1 is a block flow diagram of an embodiment of the chemical topping method based on the classification and identification of the maturity of cotton topping according to the present invention.
FIG. 2 is a flow chart of a cotton topping control system for classifying and identifying cotton topping at different maturity stages.
Fig. 3 is a schematic diagram of the operation of the cotton topping machine.
In the figure: 1. cotton plant 2, camera module 3 and topping module
Detailed Description
The present invention will be described in further detail with reference to examples and drawings, but embodiments of the present invention are not limited thereto.
Before detection, the top leaves of cotton in different maturity stages are required to be defined, wherein the size of the top leaves of cotton in maturity stage is defined, the area of the top leaves of cotton reaches more than 50% of the area of side bud leaves of cotton to be the top leaves in maturity state, otherwise, the top leaves are classified as top leaves in immature state.
Referring to fig. 1-3, the chemical topping method based on cotton topping maturity classification recognition of the present invention comprises the steps of:
s1, collecting cotton top leaf images of different maturity stages according to actual growing conditions of cotton on an operation land, wherein the cotton top leaf images comprise cotton top leaf images of different time periods, cotton top leaf images of different illumination angles and cotton top leaf images containing different top leaf types;
s2, performing data expansion in 5 modes of color enhancement, brightness enhancement, rotation, random saturation contrast sharpness and overturn on an original image by using a data enhancement method, wherein different effects of a shot cotton top leaf image, which are caused by field illumination intensity and angle change, are simulated by using the color enhancement, the brightness enhancement and the random saturation contrast sharpness, different visual angles shot by a camera are simulated by using overturn and rotation, the expanded cotton top leaf image is divided into a training set, a verification set and a test set according to the proportion of 8:1:1, a Labelimg tool is used for manual labeling and an xml type file is generated, and then the generated xml type file is stored according to the format of a PASCALVOC data set;
s3, training cotton top leaf classification and identification models based on different maturity stages, wherein the cotton top leaf classification and identification models based on different maturity stages adopt a YOLOv4 neural network model, and the following improvements are made on the basis of the YOLOv4 neural network model:
s3.1, performing feature extraction on cotton top leaf targets in different maturity stages by taking CSPDarknet53 as a main feature extraction network, and performing five-time convolution downsampling by a residual error module to realize feature extraction and screening of input information, and outputting three different scale features to the next stage;
s3.2, using Spatial Pyramid Pooling (space pyramid pooling) module and Path Aggregation Network (path aggregation network) feature pyramid network in the neck part, and performing maximum pooling treatment on the last feature layer by using pooling layers of four different scales of 1×1,3×3,5×5 and 13×13 of Spatial Pyramid Pooling (space pyramid pooling) modules; repeated up-sampling and down-sampling operations are carried out through a Path Aggregation Network (path aggregation network) module so as to realize fusion of the feature map information;
s3.3, introducing a coupled Head module into the Head module, and processing the final characteristic information again to obtain prediction frame information and confidence;
s3.4, introducing an Anchor Free method into the YOLOv4 neural network model, dividing the predicted frame information into 3 prior frames with different scales by using a label distribution method, and continuously screening out a positive sample prior frame closest to a real frame for final prediction; in the screening process, extracting the information of the positive sample prior frame, dynamically distributing the number of the positive samples by using a SimOTA method, and screening out the final positive sample prior frame by calculating the coincidence degree of the real frame and the positive sample prior frame and the type prediction accuracy data and outputting the final positive sample prior frame as a prediction result;
s4, in the process of topping cotton by the cotton topping machine, acquiring cotton top leaf images by a camera module, inputting the acquired data images into a main control module of a cotton top leaf classification and identification topping system in different maturity stages, analyzing and processing the cotton top leaf images by the main control module based on the cotton top leaf classification and identification models in different maturity stages, and outputting classification and identification results, wherein the main control module controls the cotton topping machine to accurately topping cotton.
In step S4, the trained cotton top leaf classification recognition models based on different maturity stages are deployed by adopting an online deployment mode, so that online real-time recognition of the cotton top leaf is realized.
Referring to fig. 2, in step S4, the cotton topping system for classifying and identifying cotton topping at different maturity stages includes a cotton topping machine, a camera module and a main control module, wherein the cotton topping machine includes a high-clearance vehicle and a topping module arranged on the high-clearance vehicle, and the topping module implements topping of cotton topping by spraying topping agent; the camera module is used for collecting cotton top leaf images and transmitting the collected top leaf images to the main control module, the main control module is provided with the cotton top leaf classification and identification models based on different maturity stages, the collected cotton top leaf images are analyzed and processed based on the cotton top leaf classification and identification models of different maturity stages to obtain the position information of the cotton top leaf needing to be topped, and then the topping module is controlled to spray topping agent on the cotton top leaf, wherein the camera module and the topping module are horizontally arranged in a straight line, the distance between the camera module and the topping module is 0.05m, and the camera module is positioned in front of the topping module along the movement direction of the high-clearance vehicle.
The chemical topping method based on the classification and identification of the maturity of cotton top leaves according to the invention is described in the following specific cases:
referring to fig. 1-3, the topping method based on the classification and identification of the maturity of cotton top leaves of the present invention comprises the following steps.
The first step: cultivation mode and shooting equipment for determining cotton plants
The cotton planted in a certain operation land is the robust cotton 532, the land adopts a mechanized cultivation mode with a line spacing of 76cm of one film, three lines and the like, and the shooting equipment (i.e. the camera module) is a Sony HDR-CX900E full-high definition camera.
And a second step of: collecting images of cotton top leaves at different maturity stages
(A) Collecting images of cotton top leaves in different maturity stages according to actual growth conditions of cotton on an operation land, shooting cotton top leaf images in different time periods, shooting cotton top leaf images in different illumination angles and shooting images containing different types of cotton top leaves; according to the shooting method, 3000 cotton top leaf images are shot, wherein the acquisition time is as follows: 7 am each day: 00-10: 00. 3:00-5:00, in order to simulate the shooting gesture of a camera of the cotton top leaf recognition system, the distance between the camera and the cotton top leaf is 300-500mm. Shooting angle: forward light, reverse light, light measurement. The image types include: mature top leaves, immature top leaves alone; the mature top leaf is combined with the immature top leaf, the mature top leaf is combined with the mature top leaf, and the immature top leaf is combined with the immature top leaf.
And a third step of: preparation of data sets based on top leaves of cotton at different maturity
The method comprises the steps of performing color enhancement, brightness enhancement, rotation, random saturation contrast sharpness and overturn on an original image by using a data enhancement method, wherein the color enhancement, brightness enhancement and random saturation contrast sharpness are used for simulating different effects of a shot terminal bud image along with field illumination intensity and angle change, and the overturn and rotation method is used for simulating different visual angles shot by a camera, so 12000 images are obtained after expansion, then the 12000 Zhang Dingya images are divided into 9720 training sets, 1080 verification sets and 1200 test sets according to the proportion of 8:1:1, the LabelImg tool is used for manual labeling, and the generated xml type file is saved according to the format of a PASCALVOC data set.
Fourth step: training is based on cotton top leaf classification recognition model of different maturity
The improved cotton terminal bud classification detection model of Yolov4 in different maturity stages comprises the following specific steps:
(A) Performing feature extraction on cotton terminal bud targets in different maturity stages by taking CSPDarknet53 as a trunk feature extraction network (Backbone), wherein the trunk feature extraction of a neural network model continues to use a YOLOv4 neural network model, and performing five convolution downsampling of a residual error module to realize feature extraction and screening of input information, and outputting three different scale features to the next stage;
(B) The Neck (Neck) part uses Spatial Pyramid Pooling (space pyramid pooling) module and Path Aggregation Network (path aggregation network) feature pyramid network, and the pooling of four different scales of 1×1,3×3,5×5 and 13×13 of Spatial Pyramid Pooling (space pyramid pooling) modules is utilized to carry out maximum pooling treatment on the last feature image layer, so that the shallow information of the network is fully utilized, and the connection between the feature images with different resolutions is enhanced. The up-sampling and down-sampling operations are repeatedly carried out through a Path Aggregation Network (path aggregation network) module to realize the fusion of the feature map information;
(C) Introducing a coupled Head module into the Head module, and processing the final characteristic information again to obtain prediction frame information and confidence level;
(D) An Anchor Free method is introduced into the model, the Anchor Free utilizes a label distribution method to divide a prediction frame into 3 prior frames with different scales, and a positive sample prior frame closest to a real frame is continuously screened out for final prediction. During the screening process, the positive sample prior frame information is extracted, the SimOTA method is used for dynamic positive sample quantity distribution, and the final positive sample prior frame is screened out and is used as a prediction result to be output through calculating the coincidence degree of the real frame and the positive sample prior frame, the type prediction accuracy and other data.
Fifth step: deployment is based on different maturity cotton top leaf classification recognition model
And deploying the training algorithm model in an online deployment mode according to the real-time online identification requirement of the cotton terminal buds. The model deployment can be divided into two types according to task demands, namely offline deployment and online deployment. While for online deployment, it is necessary to quickly obtain the inferred results, which is extremely sensitive to real-time. According to the method, the cotton terminal buds are required to be identified on line in real time, so that the training algorithm model is deployed in an on-line deployment mode.
Sixth step: building cotton top leaf classification, identification and topping system based on different maturity stages
(A) And loading the cotton top leaf classifying, identifying and topping systems in different maturity stages on a cotton topping machine, utilizing a camera module to collect cotton top leaf data, transmitting the collected data to a control module formed by a singlechip to calculate a result, and controlling a topping agent spray head according to the classifying and identifying result to accurately spray the topping agent. The cotton topping machine is powered by a storage battery on the high-clearance vehicle. The cotton top leaf classification and identification topping system in different maturity stages comprises a cotton topping machine, a camera module and a main control module, wherein the camera module and the main control module are arranged on the cotton topping machine; the connections between the modules and the device are shown in fig. 2;
(B) The cotton top leaf classification and identification topping system based on different maturity stages constructed in the sixth step (A) is used for completing the general design of the high-clearance vehicle according to the identification system completed in the second step to the fifth step as shown in figure 3; the high-clearance vehicle works at a certain speed of 0.1m/s, the response time of the recognition system for recognizing the image is 5s, and according to the formula L=Vt, L is the distance between the recognition camera and the spray head is 0.05m.
In addition, by improving the traditional YOLOv4 neural network model, besides comparing with the YOLOv4 model, 3 different neural network models are introduced to carry out experimental comparison with the improved YOLOv4 model, and the comparison data are shown in tables 1 and 2:
table 1: comparing the improved YOLOv4 neural network model with the traditional YOLOv4 neural network model
Table 2: comparing the improved YOLOv4 neural network model with other three neural network models
It can be seen that the detection precision mAP value of the improved YOLOv4 neural network model is 90.15%, which is respectively higher than that of the SDD, the fast-Rcnn, the Tiny-YOLOv4 neural network model by 5.95 percentage points, 13.6 percentage points and 2.17 percentage points. Specifically, in the aspect of detection of mature terminal buds, the detection accuracy AP of the improved YOLOv4 neural network model and the detection accuracy AP of the Tiny-YOLOv4 neural network model are almost equal and are above 85%; the detection precision of the Faster-Rcnn model is the lowest and is 62.42%; the improved YOLOv4 neural network model is highest in terms of the harmonic mean F1. In the aspect of immature terminal bud detection, the detection accuracy AP of the improved YOLOv4 neural network model is highest and is 94.86 percent higher than that of other models by 4 to 6 percent, and in the aspect of harmonic mean value F1, the improved YOLOv4 neural network model has the best performance and reaches 87.63 percent.
The foregoing is illustrative of the present invention and is not to be construed as limiting thereof, but rather as various changes, modifications, substitutions, combinations, and simplifications which may be made therein without departing from the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (8)
1. The chemical topping method based on the classification and identification of the mature period of the cotton top leaves is characterized by comprising the following steps of:
s1, acquiring cotton top leaf images of different maturity stages according to actual growth conditions of cotton on an operation land, wherein the steps include shooting cotton top leaf images of different time periods, shooting cotton top leaf images of different illumination angles and shooting cotton top leaf images containing different top leaf types;
s2, expanding the shot cotton top leaf image, dividing the expanded cotton top leaf image into a training set, a verification set and a test set according to the ratio of 8:1:1, manually marking by using a LabelImg tool, generating an xml type file, and storing the generated xml type file according to the format of the PASCALVOC data set;
s3, training cotton top leaf classification and identification models based on different maturity stages, wherein the cotton top leaf classification and identification models based on different maturity stages adopt a YOLOv4 neural network model, and the following improvements are made on the basis of the YOLOv4 neural network model:
s3.1, performing feature extraction on cotton top leaf targets in different maturity stages by taking CSPDarknet53 as a main feature extraction network, and performing five-time convolution downsampling by a residual error module to realize feature extraction and screening of input information, and outputting three different scale features to the next stage;
s3.2, using Spatial Pyramid Pooling module and Path Aggregation Network characteristic pyramid network in neck part, using 1×1,3×3,5×5, 13×13 pooling of Spatial Pyramid Pooling module to check the last characteristic layer for maximum pooling treatment; repeated up-sampling and down-sampling operations are carried out through a Path Aggregation Network module so as to realize fusion of the feature map information;
s3.3, introducing a coupled Head module into the Head module, and processing the final characteristic information again to obtain prediction frame information and confidence;
s3.4, introducing an Anchor Free method into the YOLOv4 neural network model, dividing the predicted frame information into 3 prior frames with different scales by using a label distribution method, and continuously screening out a positive sample prior frame closest to a real frame for final prediction; in the screening process, extracting the information of the positive sample prior frame, dynamically distributing the number of the positive samples by using a SimOTA method, and screening out the final positive sample prior frame by calculating the coincidence degree of the real frame and the positive sample prior frame and the type prediction accuracy data and outputting the final positive sample prior frame as a prediction result;
s4, in the process of topping cotton by the cotton topping machine, acquiring cotton top leaf images by a camera module, inputting the acquired data images into a main control module of a cotton top leaf classification and identification topping system of different maturity stages, analyzing and processing the cotton top leaf images by the main control module through the cotton top leaf classification and identification models of different maturity stages, and outputting classification and identification results, wherein the main control module controls the cotton topping machine to accurately topping cotton; the cotton topping machine comprises a high-clearance vehicle, a camera module arranged on the high-clearance vehicle, a topping module and a main control module, wherein the topping module adopts a mode of spraying topping agent to realize topping of cotton top leaves; the camera module is used for collecting cotton top leaf images and transmitting the collected top leaf images to the main control module, the main control module is provided with the cotton top leaf classification and identification models based on different maturity stages, the collected cotton top leaf images are analyzed and processed based on the cotton top leaf classification and identification models of different maturity stages to obtain the position information of the cotton top leaf needing topping, and then the topping module is controlled to spray topping agent to the cotton top leaf.
2. The chemical topping method based on cotton top leaf maturity categorization recognition according to claim 1, wherein in step S2, the original image is subjected to data expansion in 5 ways of color enhancement, brightness enhancement, rotation, random saturation contrast sharpness, and inversion by using a data enhancement method, wherein different effects presented by the captured cotton top leaf image as the field illumination intensity and angle change are simulated by using color enhancement, brightness enhancement, random saturation contrast sharpness, and different viewing angles captured by the camera are simulated by using inversion and rotation.
3. The chemical topping method based on cotton topping period classification and identification according to claim 1, wherein in step S4, the trained cotton topping period classification and identification model based on different maturity periods is deployed by adopting an online deployment mode, so as to realize online real-time identification of cotton topping.
4. The chemical topping method based on cotton topping maturity categorization recognition of claim 1, wherein said camera module is positioned in a horizontal straight line with said topping module with a spacing of 0.05m between the two groups, wherein said camera module is positioned in front of said topping module along the direction of movement of said high clearance vehicle.
5. The chemical topping method based on cotton topping maturity categorization recognition of claim 4, wherein said high ground clearance vehicle has a speed of movement of 0.1m/s.
6. The chemical topping method according to claim 4, wherein in step S1, the captured cotton topping image comprises a mature topping alone, an immature topping, a combination of a mature topping and a mature topping, and a combination of an immature topping and an immature topping.
7. The chemical topping method based on cotton top leaf maturity classification and identification of claim 4, wherein said main control module is a single chip microcomputer.
8. The chemical topping method based on cotton top leaf maturity categorization recognition of claim 4, wherein said camera module is a sony HDR-CX900E full high definition camera.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210486529.3A CN114916336B (en) | 2022-05-06 | 2022-05-06 | Chemical topping method based on cotton top leaf maturity stage classification and identification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210486529.3A CN114916336B (en) | 2022-05-06 | 2022-05-06 | Chemical topping method based on cotton top leaf maturity stage classification and identification |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114916336A CN114916336A (en) | 2022-08-19 |
| CN114916336B true CN114916336B (en) | 2024-03-15 |
Family
ID=82806163
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210486529.3A Active CN114916336B (en) | 2022-05-06 | 2022-05-06 | Chemical topping method based on cotton top leaf maturity stage classification and identification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114916336B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020134408A1 (en) * | 2018-12-29 | 2020-07-02 | 深圳光启空间技术有限公司 | Multi-path load-balanced asynchronous target detection method, storage medium, and processor |
| CN112200186A (en) * | 2020-10-15 | 2021-01-08 | 上海海事大学 | Car logo identification method based on improved YOLO _ V3 model |
| CN114067298A (en) * | 2021-11-18 | 2022-02-18 | 石河子大学 | Cotton tip identification method based on Faster R-CNN |
| CN114120037A (en) * | 2021-11-25 | 2022-03-01 | 中国农业科学院农业信息研究所 | Germinated potato image recognition method based on improved yolov5 model |
-
2022
- 2022-05-06 CN CN202210486529.3A patent/CN114916336B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020134408A1 (en) * | 2018-12-29 | 2020-07-02 | 深圳光启空间技术有限公司 | Multi-path load-balanced asynchronous target detection method, storage medium, and processor |
| CN112200186A (en) * | 2020-10-15 | 2021-01-08 | 上海海事大学 | Car logo identification method based on improved YOLO _ V3 model |
| CN114067298A (en) * | 2021-11-18 | 2022-02-18 | 石河子大学 | Cotton tip identification method based on Faster R-CNN |
| CN114120037A (en) * | 2021-11-25 | 2022-03-01 | 中国农业科学院农业信息研究所 | Germinated potato image recognition method based on improved yolov5 model |
Non-Patent Citations (2)
| Title |
|---|
| 化学打顶对棉花生长发育和产量的影响;袁青锋;张静;管利军;徐宇强;朱晓平;;江苏农业科学;20150525(05);全文 * |
| 基于改进YOLOv4的棉花检测算法;刘正波等;计算机系统应用;20210815;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114916336A (en) | 2022-08-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111709489B (en) | Citrus identification method based on improved YOLOv4 | |
| CN111582055B (en) | Unmanned aerial vehicle aviation pesticide application route generation method and system | |
| Tian et al. | Machine vision identification of tomato seedlings for automated weed control | |
| CN113439727B (en) | Deinsectization method, device, equipment and storage medium for greenhouse crops | |
| CN111985724B (en) | Crop yield estimation method, device, equipment and storage medium | |
| CN113920474B (en) | Internet of things system and method for intelligently supervising citrus planting situation | |
| CN113222991A (en) | Deep learning network-based field ear counting and wheat yield prediction | |
| CN110826556A (en) | Broad-spectrum crop weed identification and positioning method for improving deep learning | |
| CN114818909A (en) | Weed detection method and device based on crop growth characteristics | |
| CN111727457A (en) | Cotton crop row detection method and device based on computer vision and storage medium | |
| US20230157231A1 (en) | Methods for artificial pollination and apparatus for doing the same | |
| CN116682106A (en) | Deep learning-based intelligent detection method and device for diaphorina citri | |
| CN116129260A (en) | Forage grass image recognition method based on deep learning | |
| CN109598215A (en) | A kind of orchard Modeling Analysis System and method based on unmanned plane positioning shooting | |
| CN111832448A (en) | Disease identification method and system for grape orchard | |
| CN116128672A (en) | Model-data combined driving intelligent greenhouse fertilizer preparation method and system | |
| Zhao et al. | Transient multi-indicator detection for seedling sorting in high-speed transplanting based on a lightweight model | |
| CN113920106A (en) | Corn growth three-dimensional reconstruction and stem thickness measurement method based on RGB-D camera | |
| CN114916336B (en) | Chemical topping method based on cotton top leaf maturity stage classification and identification | |
| CN117197595A (en) | Fruit tree growth period identification method, device and management platform based on edge calculation | |
| Bortolotti et al. | A computer vision system for in-field quality evaluation: Preliminary results on peach fruit | |
| CN111191543A (en) | Rape yield estimation method | |
| CN117036926A (en) | Weed identification method integrating deep learning and image processing | |
| CN110064601A (en) | A kind of vegetable grafting seedling detection hierarchy system and stage division | |
| CN115661544A (en) | N-MobileNetXt-based spinach seedling water stress level classification system and method |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |