WO2021088700A1 - 喷头控制方法、装置、计算机设备和存储介质 - Google Patents

喷头控制方法、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2021088700A1
WO2021088700A1 PCT/CN2020/124470 CN2020124470W WO2021088700A1 WO 2021088700 A1 WO2021088700 A1 WO 2021088700A1 CN 2020124470 W CN2020124470 W CN 2020124470W WO 2021088700 A1 WO2021088700 A1 WO 2021088700A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
target
spray
target object
deconvolution
Prior art date
Application number
PCT/CN2020/124470
Other languages
English (en)
French (fr)
Inventor
张为明
Original Assignee
京东数科海益信息科技有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 京东数科海益信息科技有限公司 filed Critical 京东数科海益信息科技有限公司
Publication of WO2021088700A1 publication Critical patent/WO2021088700A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • G06T7/00—Image analysis
    • G06T7/70—Determining position or orientation of objects or cameras

Definitions

  • This application relates to the field of computer technology, in particular to a spray head control method, device, computer equipment and storage medium.
  • the sprinkler system achieves the purpose of quickly lowering the temperature of the animals by spraying water on the farmed animals.
  • the current measure is to set up multiple spray rods in the spraying hall, and fix multiple spray heads on each spray rod. When spraying is needed, turn on the main switch, and all the nozzles of the spray rods spray water. The problem with this method is that as long as there are animals, all sprinklers will spray water, even where there are no animals, causing serious waste of water resources. At the same time, it is necessary to manually trigger the spray switch to spray water, which is labor intensive.
  • this application provides a spray head control method, device, computer equipment and storage medium.
  • this application provides a nozzle control method, including:
  • the attribute information includes the spray area and spray state
  • this application provides a nozzle control device, including:
  • the image acquisition module is used to acquire an image in real time, and the image contains at least one target object;
  • the target detection module is used to input the image to the preset target detection model, and output each target object and corresponding position information;
  • the attribute information acquisition module is used to acquire the attribute information of the sprinkler, the attribute information includes the spray area and the spray state;
  • the matching module is used to match the target nozzle according to the position information and spray area of each target object;
  • the control module is used to generate an instruction for turning on the target spray head when the spray state of the target spray head is in the off state, and send the command.
  • a computer device includes a memory, a processor, and a computer program that is stored in the memory and can run on the processor, and the processor implements the following steps when the processor executes the computer program:
  • the attribute information includes the spray area and spray state
  • the computer program is executed by a processor, the following steps are implemented:
  • the attribute information includes the spray area and spray state
  • the above-mentioned spray head control method, device, computer equipment and storage medium includes: acquiring an image in real time, the image contains at least one target object; inputting the image to a preset target detection model, outputting each target object and corresponding position information; obtaining the information of the spray head Attribute information.
  • the attribute information includes spray area and spray status; match the target sprinkler according to the position information and spray area of each target object; when the spray status of the target sprinkler is off, generate an instruction to turn on the target sprinkler, Send instructions.
  • the acquired images are detected in real time through the preset target detection model, and the target object and corresponding position information in each image are output. Only the matching nozzle is turned on, that is, the spray state of the nozzle is automatically controlled according to the real-time position information corresponding to the target. There is no area where the target object is located and the corresponding nozzle is not turned on, thereby saving resources.
  • Figure 1 is an application environment diagram of a spray head control method in an embodiment
  • Figure 2 is a schematic flow chart of a method for controlling a spray head in an embodiment
  • FIG. 3 is a schematic diagram of a network structure of a bottleneck structure in an embodiment
  • FIG. 4 is a schematic structural diagram of a preset target detection model in an embodiment
  • Figure 5 is a structural block diagram of a nozzle control device in an embodiment
  • Fig. 6 is an internal structure diagram of a computer device in an embodiment.
  • Fig. 1 is an application environment diagram of a spray head control method in an embodiment.
  • the nozzle control method is applied to a nozzle control system.
  • the nozzle control system includes a photographing device 110 and a computer device 120.
  • the photographing device 110 and the computer device 120 are connected through a network.
  • the computer device 120 obtains the image of the shooting device in real time, the image contains at least one target object; inputs the image to the preset target detection model, and outputs each target object and corresponding position information; obtains the attribute information of the nozzle, the attribute information includes the spray area and Spray status; match the target sprinkler according to the position information and spray area of each target object; when the spray state of the target sprinkler is in the off state, generate an instruction for turning on the target sprinkler and send the instruction.
  • the aforementioned target detection, attribute information acquisition, nozzle matching, and instruction generation and sending processes can all be executed on the photographing device 110.
  • the photographing device 110 may be a camera or a terminal equipped with a camera. Cameras include fisheye cameras, ordinary cameras, and wide-angle cameras.
  • the computer device may be a terminal or a server, where the terminal may specifically be a desktop terminal or a mobile terminal, and the mobile terminal may specifically be at least one of a mobile phone, a tablet computer, and a notebook computer.
  • the server can be implemented as an independent server or a server cluster composed of multiple servers.
  • a method for controlling a spray head is provided.
  • the method is mainly applied to the photographing device 110 (or the computer device 120) in FIG. 1 as an example.
  • the nozzle control method specifically includes the following steps:
  • step S201 an image is acquired in real time.
  • the image contains at least one target object.
  • Step S202 Input an image to a preset target detection model, and output each target object and corresponding position information.
  • the image is an image taken by each photographing device, and different photographing devices are used to monitor different areas.
  • the target object refers to the animals raised in the farm, including but not limited to cattle and pigs.
  • the same image can contain multiple target objects, and the multiple target objects contained can be of the same type or of different types.
  • the types of target objects contained in multiple different images may be the same or different.
  • the preset target detection model is a convolutional segmentation network model trained on a large number of images carrying target object types and corresponding location information.
  • convolutional segmentation network models include but are not limited to ENet, FCN (Fully Convolutional Networks) and other networks.
  • the ENet network is mainly divided into two modules: encoding module and decoding module.
  • the entire network structure is composed of bottleneck.
  • the bottleneck structure is shown in Figure 3 and is divided into three convolutional layers: the first 1x1 convolution is used to reduce features Dimensions; the second is the main convolution layer, which can be ordinary convolution, hole convolution, and asymmetric convolution with a convolution kernel of mx1 or 1xm; the third 1x1 convolution is used to expand the dimension.
  • Step S203 Obtain the attribute information of the spray head.
  • the attribute information includes spray area and spray state.
  • the sprinkler attribute information includes the sprinkler's sprinkler area, sprinkler status, spray flow rate, etc.
  • the sprinkler area is related to the structure and installation height of the sprinkler itself, and the sprinkler area of each sprinkler can overlap or not. overlapping.
  • the spray state includes, but is not limited to, an open state, a closed state, and so on.
  • the method for acquiring the attribute information of the print head can be customized according to requirements, such as real-time collection of the attribute information of the print head, or a data collection instruction for collecting the attribute information of the print head is generated after the target object is detected.
  • step S204 the target spray head is matched according to the position information of each target object and the spray area.
  • step S205 when the spray state of the target spray head is in the closed state, an instruction for turning on the target spray head is generated, and the instruction is sent.
  • the position information of the target object includes the center position information of the target, the area position information of the area where it is located, and so on.
  • the nozzles are matched according to the position information of each target object and the position information corresponding to the spray area.
  • the same target object may be located in the spray area of multiple nozzles. Therefore, after determining at least one nozzle according to the position information, the position needs to be further determined. Confirmation, such as judging the proportion of the target object in each spray area, where the proportion can be expressed according to the proportion of the target object located in the spray area of each nozzle. It is also possible to use the position of the center point of the target object to determine which spray area the target object is located in.
  • the spray state of the target nozzle is closed, which means that the target object is located in the spray area corresponding to the target nozzle, and the nozzle does not spray the corresponding liquid, so an instruction for turning on the nozzle is generated, and the instruction is sent to control the nozzle
  • the controller executes the instruction to realize the spraying of liquid.
  • the above-mentioned spray head control method includes: acquiring an image in real time, the image contains at least one target object; inputting the image to a preset target detection model, outputting each target object and corresponding position information; acquiring attribute information of the spray head, the attribute information including the spray area And spray status; match the target sprinkler according to the position information and spray area of each target object; when the spray state of the target sprinkler is in the off state, generate an instruction to turn on the target sprinkler and send the instruction.
  • the acquired images are detected in real time through the preset target detection model, and the target object and corresponding position information in each image are output. Only the matching nozzle is turned on, that is, the spray state of the nozzle is automatically controlled according to the real-time position information corresponding to the target. There is no area where the target object is located and the corresponding nozzle is not turned on, thereby saving resources.
  • matching the target spray head according to the position information of the target object and the spray area includes: searching for candidate spray areas according to the position information of the target object to obtain at least one candidate spray area; calculating that the target object is located in each candidate spray area The area ratio of the area; the candidate spray area corresponding to the area ratio greater than the preset ratio is used as the target spray area; the nozzle corresponding to the target spray area is used as the target nozzle.
  • the candidate spray area refers to a spray area that overlaps with the area where the target object is located, where the candidate spray area may include one or more.
  • the target object is located in two spray areas, and the overlap area between the target object area and the two spray areas is greater than the preset value, if the two nozzles are not turned on, an instruction to turn on the two nozzles will be generated, by executing Control the opening of the nozzle.
  • the target nozzle is determined according to the ratio of the area of the overlapping area to the area where the target object is located, and the target nozzle can be positioned more accurately.
  • the above nozzle control method further includes: generating a preset target detection model, wherein generating the preset target detection model includes:
  • Step S301 Obtain multiple training images.
  • each training image carries tag information
  • the tag information includes the target object and corresponding location information.
  • Step S302 input each training image to the initial target detection model, and output the predicted label of each training image.
  • the predicted label includes predicted target object and predicted location information.
  • step S303 the model parameters of the initial target detection model are updated according to the label information of each training image and the corresponding predicted label, until the initial target detection model meets the preset convergence condition, and the preset target detection model is obtained.
  • the training image contains the target object, and the image carries the position information of each target object, the type information of the target object, and so on.
  • the type of the target object may be the same or different. If it is the same type, the type may not be marked, and if it contains multiple types, the corresponding type may be marked.
  • the predicted location information includes location center point coordinates, area coordinates, and so on. According to the predicted label and label of each training image, it is judged whether the model meets the preset convergence condition.
  • the preset target detection model is obtained. Otherwise, the initial target detection model is updated according to the difference degree of each predicted label and label.
  • the method of updating model parameters is a common machine learning parameter updating method.
  • the loss function corresponding to the preset convergence condition can be a common machine learning loss function, and minimizing the loss function makes the initial target detection model converge.
  • the preset target detection model includes an encoding module and a decoding module.
  • the network structure corresponding to the encoding module is the same as the first 15 modules in the ENet network.
  • the decoding module includes a deconvolution unit and a full convolution unit. The input of the product unit is the output of the deconvolution unit.
  • the decoding module is used to extract the features of the image
  • the deconvolution unit is used to use the deconvolution operation on the features extracted by the decoding module to increase the size of the feature
  • the full convolution unit is used to implement detection based on the up-sampled data , Get the target object and corresponding location information.
  • the encoding module includes a down-sampling unit and a convolution unit, inputting an image to a preset target detection model, and outputting each target object and corresponding position information, including: inputting an image to the down-sampling unit to perform the next sampling operation, Get the down-sampling feature of the image; the down-sampling feature of the input image is sent to the convolution unit to perform the convolution operation to obtain the convolution feature of the image; the convolution feature of the input image is sent to the deconvolution unit to perform the deconvolution operation to obtain the image Deconvolution feature, input the deconvolution feature of the image to the full convolution unit to perform the full convolution operation, and output the target object of the image and the corresponding position information.
  • the down-sampling unit is used to reduce the amount of image data, thereby speeding up data processing efficiency. If you perform a down-sampling operation, the 100*100 image becomes a 25*25 image.
  • the convolution unit is used to extract features in the image, and the extracted features are used to describe the target object.
  • the convolution unit includes at least one hole convolution unit, at least one asymmetric convolution unit, at least one ordinary convolution unit, and so on.
  • the hole convolution unit is used for hole convolution to reduce the amount of calculation and increase the receptive field.
  • the weight of the convolutional layer has considerable redundancy, and the use of asymmetric convolutional layer cascade can reduce the amount of calculation.
  • the network structure of the deconvolution unit is that the first deconvolution subunit is followed by two bottleneck structures, the two bottleneck structures are followed by the second deconvolution subunit, and the second deconvolution subunit is followed by Then a bottleneck structure.
  • FIG. 4 is a network structure diagram of a preset target detection model, including two parts of encoding and decoding.
  • the step of generating a preset target detection model includes:
  • the original image and the label image are input to the improved ENet segmentation network.
  • the improved ENet segmentation network first uses the initialization unit to reduce the space size, and merges the maximum pooling feature and the convolutional feature.
  • the feature map is restored to the same size as the original image. That is, two deconvolution and one full convolution layer are used for upsampling, and the feature resolution is restored to obtain the decoded feature.
  • the decoded features are classified at the pixel level, and the loss loss function is calculated according to the classification result and the label image, and the loss function is continuously propagated back to the network convergence to obtain a preset target detection model.
  • the spray bar After segmenting the target object, it is necessary to determine the spray bar to which the target object at the current position belongs. For example, the center line of two adjacent nozzles is the boundary of the spray area.
  • Use LabelMe to mark the spray area of each nozzle, and use the image processing method to calculate the area of the cow body falling into each spray area and the spray area. When the proportion is greater than a certain threshold, the corresponding nozzle performs spraying.
  • Fig. 2 is a schematic flow chart of a spray head control method in an embodiment. It should be understood that although the various steps in the flowchart of FIG. 2 are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless there is a clear description in this article, there is no strict order for the execution of these steps, and these steps can be executed in other orders. Moreover, at least part of the steps in FIG. 2 may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution of these sub-steps or stages The sequence is not necessarily performed sequentially, but may be performed alternately or alternately with at least a part of other steps or sub-steps or stages of other steps.
  • a spray device 200 including:
  • the image acquisition module 201 is used to acquire an image in real time, and the image contains at least one target object.
  • the target detection module 202 is used for inputting an image to a preset target detection model, and outputting each target object and corresponding position information.
  • the attribute information acquisition module 203 is used to acquire the attribute information of the spray head.
  • the attribute information includes the spray area and the spray state.
  • the matching module 204 is used to match the target spray head according to the position information and spray area of each target object.
  • the control module 205 is used to generate and send an instruction for turning on the target nozzle when the spray state of the target nozzle is in the off state.
  • the matching module 204 is specifically configured to search for candidate spray areas according to the position information of the target object to obtain at least one candidate spray area, and calculate the area occupied by the target object in each of the candidate spray areas.
  • the candidate spray area corresponding to the area accounted for greater than the preset account ratio is used as the target spray area, and the spray head corresponding to the target spray area is used as the target spray head.
  • the above-mentioned nozzle control device 200 further includes:
  • the model generation module is used to obtain multiple training images.
  • Each training image carries label information.
  • the label information includes the target object and corresponding position information. It inputs each training image to the initial target detection model, outputs the predicted label of each training image, and predicts
  • the label includes the predicted target object and predicted position information; the model parameters of the initial target detection model are updated according to the label information of each training image and the corresponding predicted label, until the initial target detection model meets the preset convergence conditions, and the preset target detection model is obtained.
  • Fig. 6 shows an internal structure diagram of a computer device in an embodiment.
  • the computer device may specifically be the photographing device 110 (or the computer device 120) in FIG. 1.
  • the computer equipment is connected to the processor, memory, network interface, input device and display screen through the system bus.
  • the memory includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium of the computer device stores an operating system and may also store a computer program.
  • the processor can realize the spray head control method.
  • a computer program may also be stored in the internal memory, and when the computer program is executed by the processor, the processor can execute the nozzle control method.
  • the display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen. It can be an external keyboard, touchpad, or mouse.
  • FIG. 6 is only a block diagram of part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied.
  • the specific computer device may Including more or fewer parts than shown in the figure, or combining some parts, or having a different arrangement of parts.
  • the spray device provided in the present application can be implemented in the form of a computer program, and the computer program can be run on the computer device as shown in FIG. 6.
  • the memory of the computer equipment can store various program modules that make up the spray device, for example, the image acquisition module 201, the target detection module 202, the attribute information acquisition module 203, the matching module 204, and the control module 205 shown in FIG. 5.
  • the computer program composed of each program module causes the processor to execute the steps in the nozzle control device of each embodiment of the application described in this specification.
  • the computer device shown in FIG. 6 can acquire images in real time through the image acquisition module 201 in the shower device shown in FIG. 5, and the images contain at least one target object.
  • the computer device can execute the input image to the preset target detection model through the target detection module 202, and output each target object and corresponding position information.
  • the computer device can acquire the attribute information of the spray head through the attribute information acquisition module 203, and the attribute information includes the spray area and the spray state.
  • the computer device can match the target spray head according to the position information and spray area of each target object through the matching module 204.
  • the computer device can execute the control module 205 to generate an instruction for turning on the target sprinkler head when the spray state of the target sprinkler head is in the off state, and send the instruction.
  • a computer device including a memory, a processor, and a computer program stored in the memory and capable of running on the processor.
  • the processor executes the computer program, the following steps are implemented: real-time acquisition of images, Contains at least one target object; input the image to the preset target detection model, output each target object and corresponding position information; obtain the attribute information of the spray head, the attribute information includes the spray area and spray state; according to the position information of each target object and The spray area matches the target nozzle; when the spray state of the target nozzle is off, an instruction for turning on the target nozzle is generated and the instruction is sent.
  • matching the target spray head according to the position information of the target object and the spray area includes: searching for candidate spray areas according to the position information of the target object to obtain at least one candidate spray area; calculating that the target object is located in each candidate spray area The area ratio of the area; the candidate spray area corresponding to the area ratio greater than the preset ratio is used as the target spray area; the nozzle corresponding to the target spray area is used as the target nozzle.
  • the processor further implements the following steps when executing the computer program: generating a preset target detection model, including: acquiring a plurality of training images, each training image carries label information, and the label information includes the target object and corresponding position information ; Input each training image to the initial target detection model, output the predicted label of each training image, the predicted label includes predicted target object and predicted position information; update the model parameters of the initial target detection model according to the label information of each training image and the corresponding predicted label , Until the initial target detection model meets the preset convergence condition, the preset target detection model is obtained.
  • the preset target detection model includes an encoding module and a decoding module.
  • the network structure corresponding to the encoding module is the same as the first 15 modules in the ENet network.
  • the decoding module includes a deconvolution unit and a full convolution unit. The input of the product unit is the output of the deconvolution unit.
  • the encoding module includes a down-sampling unit and a convolution unit, inputting an image to a preset target detection model, and outputting each target object and corresponding position information, including: inputting an image to the down-sampling unit to perform the next sampling operation, Get the down-sampling feature of the image; the down-sampling feature of the input image is sent to the convolution unit to perform the convolution operation to obtain the convolution feature of the image; the convolution feature of the input image is sent to the deconvolution unit to perform the deconvolution operation to obtain the image Deconvolution feature: The deconvolution feature of the input image is sent to the full convolution unit to perform the full convolution operation, and the target object of the image and the corresponding position information are output.
  • the network structure of the deconvolution unit is that the first deconvolution subunit is followed by two bottleneck structures, the two bottleneck structures are followed by the second deconvolution subunit, and the second deconvolution subunit is followed by Then a bottleneck structure.
  • a computer-readable storage medium is provided, and a computer program is stored thereon.
  • the computer program is executed by a processor, the following steps are realized: acquiring an image in real time, and the image contains at least one target object; Set up the target detection model, output each target object and corresponding position information; obtain the attribute information of the nozzle, which includes the spray area and spray state; match the target nozzle according to the position information and spray area of each target object; when the target nozzle When the spray state of is in the off state, an instruction to turn on the target nozzle is generated and the instruction is sent.
  • matching the target spray head according to the position information of the target object and the spray area includes: searching for candidate spray areas according to the position information of the target object to obtain at least one candidate spray area; calculating that the target object is located in each candidate spray area The area ratio of the area; the candidate spray area corresponding to the area ratio greater than the preset ratio is used as the target spray area; the nozzle corresponding to the target spray area is used as the target nozzle.
  • the following steps are also implemented: generating a preset target detection model, including: acquiring a plurality of training images, each training image carries label information, and the label information includes the target object and the corresponding position Information; input each training image to the initial target detection model, and output the predicted label of each training image.
  • the predicted label includes the predicted target object and predicted position information; the model of the initial target detection model is updated according to the label information of each training image and the corresponding predicted label Parameters until the initial target detection model meets the preset convergence condition, and the preset target detection model is obtained.
  • the preset target detection model includes an encoding module and a decoding module.
  • the network structure corresponding to the encoding module is the same as the first 15 modules in the ENet network.
  • the decoding module includes a deconvolution unit and a full convolution unit. The input of the product unit is the output of the deconvolution unit.
  • the encoding module includes a down-sampling unit and a convolution unit, inputting an image to a preset target detection model, and outputting each target object and corresponding position information, including: inputting an image to the down-sampling unit to perform the next sampling operation, Get the down-sampling feature of the image; the down-sampling feature of the input image is sent to the convolution unit to perform the convolution operation to obtain the convolution feature of the image; the convolution feature of the input image is sent to the deconvolution unit to perform the deconvolution operation to obtain the image Deconvolution feature: The deconvolution feature of the input image is sent to the full convolution unit to perform the full convolution operation, and the target object of the image and the corresponding position information are output.
  • the network structure of the deconvolution unit is that the first deconvolution subunit is followed by two bottleneck structures, the two bottleneck structures are followed by the second deconvolution subunit, and the second deconvolution subunit is followed by Then a bottleneck structure.
  • Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
  • SRAM static RAM
  • DRAM dynamic RAM
  • SDRAM synchronous DRAM
  • DDRSDRAM double data rate SDRAM
  • ESDRAM enhanced SDRAM
  • SLDRAM synchronous chain Channel
  • memory bus Radbus direct RAM
  • RDRAM direct memory bus dynamic RAM
  • RDRAM memory bus dynamic RAM

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)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

本申请涉及一种喷头控制方法、装置、计算机设备和存储介质。所述方法包括：实时获取拍摄设备的图像，图像中包含至少一个目标对象；输入图像至预设目标检测模型，输出各个目标对象和对应的位置信息；获取喷头的属性信息，属性信息包括喷淋区域和喷淋状态；根据各个目标对象的位置信息和喷淋区域匹配目标喷头；当目标喷头的喷淋状态为关闭状态时，生成用于开启目标喷头的指令，发送指令。通过预设目标检测模型实时检测获取到的图像，输出各个图像中的目标对象和对应的位置信息，只有匹配的喷头才开启，即根据目标对应的实时位置信息自动控制喷头的喷淋状态，对应不存在目标对象所在区域不开启对应的喷头，从而节约资源。

Description

喷头控制方法、装置、计算机设备和存储介质
本申请要求于2019年11月05日提交中国专利局、申请号为201911073264.9、发明名称为“喷头控制方法、装置、计算机设备和存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及计算机技术领域，尤其涉及一种喷头控制方法、装置、计算机设备和存储介质。
背景技术
随着我国经济的不断发展、人们生活水平的不断提高，人们的消费观念和消费水平也有了很大的转变与提升。在这个契机下，养殖行业也得到了快速发展，养殖规模不断壮大。
然而，夏季高温对于养殖场来说是一项非常严峻的考验，天气炎热非常容易导致养殖的动物出现热应激现象。热应激会影响养殖的动物的生产性能、损害养殖动物的身体等等，从而损害养殖场的经济效益。由于我国幅员辽阔，从北方到南方，从东北到西北，夏季热应激期间热应激不断加剧。
为了给养殖的动物进行降温，目前的主要技术是采用喷淋系统。喷淋系统通过在养殖的动物上喷水，达到快速降低动物温度的目的。
目前采取的措施是在进入喷淋厅内设置多根喷淋杆，每根喷淋杆上固定多个喷头，需要喷淋时打开总开关，所有喷淋杆的喷头全部喷水。这种方式带来的问题是只要有动物，全部喷头都会喷水，包括没有动物的地方也会喷，造成严重的水资源浪费。同时，需要人工触发 喷淋开关喷水，耗费人力。
发明内容
为了解决上述技术问题，本申请提供了一种喷头控制方法、装置、计算机设备和存储介质。
第一方面，本申请提供了一种喷头控制方法，包括：
实时获取图像，图像中包含至少一个目标对象；
输入图像至预设目标检测模型，输出各个目标对象和对应的位置信息；
获取喷头的属性信息，属性信息包括喷淋区域和喷淋状态；
根据各个目标对象的位置信息和喷淋区域匹配目标喷头；
当目标喷头的喷淋状态为关闭状态时，生成用于开启目标喷头的指令，发送指令。
第二方面，本申请提供了一种喷头控制装置，包括：
图像获取模块，用于实时获取图像，图像中包含至少一个目标对象；
目标检测模块，用于输入图像至预设目标检测模型，输出各个目标对象和对应的位置信息；
属性信息获取模块，用于获取喷头的属性信息，属性信息包括喷淋区域和喷淋状态；
匹配模块，用于根据各个目标对象的位置信息和喷淋区域匹配目标喷头；
控制模块，用于当目标喷头的喷淋状态为关闭状态时，生成用于开启目标喷头的指令，发送所述指令。
一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现以下步骤：
实时获取图像，图像中包含至少一个目标对象；
输入图像至预设目标检测模型，输出各个目标对象和对应的位置信息；
获取喷头的属性信息，属性信息包括喷淋区域和喷淋状态；
根据各个目标对象的位置信息和喷淋区域匹配目标喷头；
当目标喷头的喷淋状态为关闭状态时，生成用于开启目标喷头的指令，发送指令。
一种计算机可读存储介质，其上存储有计算机程序，计算机程序被处理器执行时实现以下步骤：
实时获取图像，图像中包含至少一个目标对象；
输入图像至预设目标检测模型，输出各个目标对象和对应的位置信息；
获取喷头的属性信息，属性信息包括喷淋区域和喷淋状态；
根据各个目标对象的位置信息和喷淋区域匹配目标喷头；
当目标喷头的喷淋状态为关闭状态时，生成用于开启目标喷头的指令，发送指令。
上述喷头控制方法、装置、计算机设备和存储介质，方法包括：实时获取图像，图像中包含至少一个目标对象；输入图像至预设目标检测模型，输出各个目标对象和对应的位置信息；获取喷头的属性信息，属性信息包括喷淋区域和喷淋状态；根据各个目标对象的位置信息和喷淋区域匹配目标喷头；当目标喷头的喷淋状态为关闭状态时， 生成用于开启目标喷头的指令，发送指令。通过预设目标检测模型实时检测获取到的图像，输出各个图像中的目标对象和对应的位置信息，只有匹配的喷头才开启，即根据目标对应的实时位置信息自动控制喷头的喷淋状态，对应不存在目标对象所在区域不开启对应的喷头，从而节约资源。
附图说明
此处的附图被并入说明书中并构成本说明书的一部分，示出了符合本申请的实施例，并与说明书一起用于解释本申请的原理。
为了更清楚地说明本申请实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，对于本领域普通技术人员而言，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
图1为一个实施例中喷头控制方法的应用环境图；
图2为一个实施例中喷头控制方法的流程示意图；
图3为一个实施例中bottleneck结构的网络结构示意图；
图4为一个实施例中预设目标检测模型的结构示意图；
图5为一个实施例中喷头控制装置的结构框图；
图6为一个实施例中计算机设备的内部结构图。
具体实施方式
为使本申请实施例的目的、技术方案和优点更加清楚，下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本申请的一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动的前提下所获得的所有其他实施例，都属于本申请 保护的范围。
图1为一个实施例中喷头控制方法的应用环境图。参照图1，该喷头控制方法应用于喷头控制系统。该喷头控制系统包括拍摄设备110和计算机设备120。拍摄设备110和计算机设备120通过网络连接。计算机设备120实时获取拍摄设备的图像，图像中包含至少一个目标对象；输入图像至预设目标检测模型，输出各个目标对象和对应的位置信息；获取喷头的属性信息，属性信息包括喷淋区域和喷淋状态；根据各个目标对象的位置信息和喷淋区域匹配目标喷头；当目标喷头的喷淋状态为关闭状态时，生成用于开启目标喷头的指令，发送指令。
上述目标检测、属性信息获取、喷头匹配和指令生成和发送过程均可以在拍摄设备110上执行。
拍摄设备110可以为摄像头或搭载有摄像头的终端。摄像头包括鱼眼摄像头、普通摄像头、广角摄像头等。计算机设备可以为终端或服务器，其中终端具体可以是台式终端或移动终端，移动终端具体可以手机、平板电脑、笔记本电脑等中的至少一种。服务器可以用独立的服务器或者是多个服务器组成的服务器集群来实现。
如图2所示，在一个实施例中，提供了一种喷头控制方法。本实施例主要以该方法应用于上述图1中的拍摄设备110(或计算机设备120)来举例说明。参照图2，该喷头控制方法具体包括如下步骤：
步骤S201，实时获取图像。
在本实施例中，图像中包含至少一个目标对象。
步骤S202，输入图像至预设目标检测模型，输出各个目标对象和对应的位置信息。
具体地，图像为各个拍摄设备拍摄的图像，其中不同的拍摄设备用于监控不同的区域。目标对象是指养殖场中养殖的动物，动物包括 但不限于牛、猪等。同一张图像中可以包含多个目标对象，包含的多个目标对象可以为相同类型，也可以为不同类型。多张不同的图像中包含的目标对象的类型可以相同也可以不同。
预设目标检测模型是根据大量的携带目标对象类型和对应的位置信息的图像训练得到的卷积分割网络模型。其中卷积分割网络模型包括但不限于ENet、FCN(Fully Convolutional Networks)等等网络。其中ENet网络主要分为编码模块和解码模块两个模块，其中整个网络结构由bottleneck组成，其中bottleneck结构如图3所示，分为三个卷积层：第一个1x1卷积用来降低特征维度；第二个为主卷积层，可以为普通卷积、空洞卷积和卷积核为mx1或1xm的非对称卷积；第三个1x1卷积用来扩张维度。若bottleneck为下采样，则在主分支中加入一个最大池化层(MaxPooling)，同时卷积核大小为2x2、步长为2的卷积代替第一个1x1卷积，对激活值进行padding操作，保证两个分支的数据在融合时具有相同的特征尺寸。
步骤S203，获取喷头的属性信息。
在本实施例中，属性信息包括喷淋区域和喷淋状态。
具体地，喷头属性信息包括喷头的喷淋区域、喷淋状态、喷淋的流量等级等等，其中喷淋区域与喷头本身结构和设置高度等相关，且各个喷头的喷淋区域可以重叠或不重叠。喷淋状态包括但不限于开启状态和关闭状态等等。喷头的属性信息的获取方法可以根据需求自定义设置，如实时采集喷头的属性信息，或在检测到目标对象后生成用于采集喷头的属性信息的数据采集指令。
步骤S204，根据各个目标对象的位置信息和喷淋区域匹配目标喷头。
步骤S205，当目标喷头的喷淋状态为关闭状态时，生成用于开启目标喷头的指令，发送指令。
具体地，目标对象的位置信息包括目标的中心位置信息、所在区域的区域位置信息等等。根据各个目标对象的位置信息和喷淋区域对应的位置信息对喷头进行匹配，同一目标对象可能位于多个喷头的喷淋区域，故在根据位置信息确定至少一个喷头后，需要对位置进行进一步的确认，如判断目标对象在各个喷淋区域的所占有的比例，其中占有比例可以根据目标对象位于各个喷头的喷淋区域的面积的占比表示。也可以采用位于目标对象的中心点的位置判断目标对象位于哪个喷淋区域等。还可以是中心点位置和区域占比同时考虑等等。目标喷头的喷淋状态为关闭状态，表示目标对象位于目标喷头对应的喷淋区域，而该喷头未喷出对应的液体，故生成用于开启该喷头的指令，发送该指令，以使控制喷头的控制器执行该指令，实现对液体的喷洒。
上述喷头控制方法，包括：实时获取图像，图像中包含至少一个目标对象；输入图像至预设目标检测模型，输出各个目标对象和对应的位置信息；获取喷头的属性信息，属性信息包括喷淋区域和喷淋状态；根据各个目标对象的位置信息和喷淋区域匹配目标喷头；当目标喷头的喷淋状态为关闭状态时，生成用于开启目标喷头的指令，发送指令。通过预设目标检测模型实时检测获取到的图像，输出各个图像中的目标对象和对应的位置信息，只有匹配的喷头才开启，即根据目标对应的实时位置信息自动控制喷头的喷淋状态，对应不存在目标对象所在区域不开启对应的喷头，从而节约资源。
在一个实施例中，根据目标对象的位置信息和喷淋区域匹配目标喷头，包括：根据目标对象的位置信息查找候选喷淋区域，得到至少一个候选喷淋区域；计算目标对象位于各个候选喷淋区域的面积占比；将面积占比大于预设占比对应的候选喷淋区域，作为目标喷淋区域；将目标喷淋区域对应的喷头作为目标喷头。
具体地，候选喷淋区域是指与目标对象所在的区域存在重叠的喷淋区域，其中，候选喷淋区域可以包含一个或多个。计算目标对象所 在的区域与各个候选喷淋区域的重叠区域的面积，计算重叠各个重叠区域面积与目标对象所在的区域的面积的比值，即面积占比，其中面积占比可以通过像素点占比得到。判断该比值是否大于预设占比，当该比值大于预设占比的候选喷淋区域作为目标喷淋区域，将目标喷淋区域对应的喷头作为目标喷头。其中目标喷头可以为一个或多个。如目标对象位于两个喷淋区域内，且目标对象所在区域与两个喷淋区域的重叠区域都大于预设值，若两个喷头都未开启，则生成开启两个喷头的指令，通过执行控制喷头的开启。根据重叠区域的区域面积和目标对象的位于的区域面积的比值确定目标喷头，能够较为准确的定位目标喷头。
在一个实施例中，上述喷头控制方法，还包括：生成预设目标检测模型，其中生成预设目标检测模型，包括：
步骤S301，获取多个训练图像。
在本具体实施例中，各个训练图像中携带标签信息，标签信息包括目标对象和对应的位置信息。
步骤S302，输入各个训练图像至初始目标检测模型，输出各个训练图像的预测标签。
在本具体实施例中，预测标签包括预测目标对象和预测位置信息。
步骤S303，根据各个训练图像的标签信息和对应的预测标签更新初始目标检测模型的模型参数，直至初始目标检测模型满足预设收敛条件，得到预设目标检测模型。
具体地，训练图像中包含目标对象，且图像携带各个目标对象的位置信息和目标对象的类型信息等。其中目标对象的类型可以为相同也可以不相同，若为相同的类型可以不标注类型，若包含多个类型时，则可以标注对应的类型。将图像输入初始目标检测模型，通过初始目 标检测模型对训练图像进行特征提取，根据提取到的特征识别目标对象和检测目标对象所在的位置，即得到训练图像对应的预测目标对象和对应的预测位置信息。其中预测位置信息包括位置中心点坐标、区域坐标等等。根据各个训练图像的预测标签和标签判断模型是否满足预设收敛条件，若初始目标检测模型满足预设收敛条件，得到预设目标检测模型，反之，则根据各个预测标签和标签的差异度更新初始目标检测模型的模型参数，直至更新了模型参数的初始目标检测模型满足预设收敛条件，得到预设目标检测模型。其中更新模型参数的方法为常见的机器学习的参数更新方法。其中预设收敛条件对应的损失函数可以为常见的机器学习的损失函数，最小化损失函数使得初始目标检测模型收敛。
在一个实施例中，预设目标检测模型包括编码模块和解码模块，编码模块对应的网络结构与ENet网络中的前15个模块相同，解码模块包括反卷积单元和全卷积单元，全卷积单元的输入为反卷积单元的输出。
具体地，解码模块用于提取图像的特征，反卷积单元是用于对解码模块提取的特征采用反卷积操作增大特征的尺寸，全卷积单元用于根据上采样后的数据实现检测，得到目标对象和对应的位置信息。
在一个实施例中，编码模块包括下采样单元和卷积单元，输入图像至预设目标检测模型，输出各个目标对象和对应的位置信息，包括：输入图像至下采样单元执行下次采样操作，得到图像的下采样特征；输入图像的下采样特征至卷积单元执行卷积操作，得到图像的卷积特征；输入图像的卷积特征至反卷积单元执行反卷积操作，得到各个图像的反卷积特征，输入图像的反卷积特征至全卷积单元执行全卷积操作，输出图像的目标对象和对应的位置信息。
具体地，下采样单元用于减少图像的数据量，从而加快数据处理 效率。如执行下采样操作，将100*100的图像变成25*25的图像。卷积单元用于提取图像中的特征，提取出的特征用于描述目标对象。其中卷积单元包括至少一个空洞卷积单元、至少一个不对称卷积单元和至少一个普通卷积单元等。其中空洞卷积单元是用于空洞卷积减小计算量和增大感受野。卷积层权重有相当大的冗余，采用不对称卷积层级联可以缩小计算量。
在一个实施例中，反卷积单元的网络结构为第一反卷积子单元后接两个bottleneck结构，两个bottleneck结构后接第二反卷积子单元，第二反卷积子单元后接一个bottleneck结构。
在一个具体的实施例中，参照图3和图4，图4为一个预设目标检测模型的网络结构图，包括编码和解码两个部分。预设目标检测模型为改进的ENet分割网络，其中初始化层用于实现下采样，下采样bottleneck1.0和4个常规卷积bottleneck1.x，下采样bottleneck2.0，常规卷积bottleneck2.1，超参rate＝2的空洞卷积bottleneck2.2，不对称卷积bottleneck2.3，超参rate＝4的空洞卷积bottleneck2.4，常规卷积bottleneck2.5，超参rate＝8的空洞卷积bottleneck2.6，不对称卷积bottleneck2.7，超参rate＝16的空洞卷积bottleneck2.8，反卷积bottleneck3.0，常规卷积bottleneck3.1和bottleneck3.2，反卷积bottleneck4.1，常规卷积bottleneck4.1，最后一层为全卷积。
在一个具体地实施例中，生成预设目标检测模型的步骤，包括：
获取摄像头拍摄的包含目标对象的原始图像，采用PixelAnnotationTool对原始图像中的目标对象所在区域进行标记，生成标签图像；
将原始图像和标签图像输入改进的ENet分割网络，改进的ENet分割网络先用初始化单元缩小空间尺寸，合并最大池化特征和卷积后特征。
通过一系列的bottleneck网络对牛棚图像进行编码特征提取，将图像下采样了8倍。其中考虑到卷积层权重有相当大的冗余，可以用5x1和1x5的两个不对称卷积层级联来缩小计算量。同时在维持特征图的分辨率的情况下，引入空洞卷积减小计算量、增大感受野。
编码之后为解码过程，将特征图还原到与原始图像相同的大小。即采用两个反卷积和一个完全卷积层进行上采样，还原特征分辨率得到解码特征。对解码特征进行像素级别的分类，并根据分类结果和标签图像计算loss损失函数，不断反向传播损失函数至网络收敛，得到预设目标检测模型。
将拍摄到的图像输入预设目标检测模型，即可到分割出的目标对象的位置信息。
在分割出目标对象之后，需确定当前位置的目标对象归属的喷淋杆。如两根相邻喷头的中线为喷淋区域的界限，使用LabelMe标记出各个喷头的喷淋区域，用图像处理的方法计算出落入每个喷淋区域的牛体面积与该喷淋区域的占比，当占比大于一定阈值，相应的喷头执行喷淋。
上述根据目标对象的位置实时喷水，无人化喷淋，节省大量喷淋液体和人力资源。
图2为一个实施例中喷头控制方法的流程示意图。应该理解的是，虽然图2的流程图中的各个步骤按照箭头的指示依次显示，但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明，这些步骤的执行并没有严格的顺序限制，这些步骤可以以其它的顺序执行。而且，图2中的至少一部分步骤可以包括多个子步骤或者多个阶段，这些子步骤或者阶段并不必然是在同一时刻执行完成，而是可以在不同的时刻执行，这些子步骤或者阶段的执行顺序也不必然是依次进行，而是可以与其它步骤或者其它步骤的子步骤或者阶段 的至少一部分轮流或者交替地执行。
在一个实施例中，如图5所示，提供了一种喷淋装置200，包括：
图像获取模块201，用于实时获取图像，图像中包含至少一个目标对象。
目标检测模块202，用于输入图像至预设目标检测模型，输出各个目标对象和对应的位置信息。
属性信息获取模块203，用于获取喷头的属性信息，属性信息包括喷淋区域和喷淋状态。
匹配模块204，用于根据各个目标对象的位置信息和喷淋区域匹配目标喷头。
控制模块205，用于当目标喷头的喷淋状态为关闭状态时，生成用于开启目标喷头的指令，发送指令。
在一个实施例中，匹配模块204具体用于根据所述目标对象的位置信息查找候选喷淋区域，得到至少一个候选喷淋区域，计算所述目标对象位于各个所述候选喷淋区域的面积占比，将所述面积占比大于预设占比对应的所述候选喷淋区域，作为目标喷淋区域，将所述目标喷淋区域对应的喷头作为所述目标喷头。
在一个实施例中，上述喷头控制装置200，还包括：
模型生成模块，用于获取多个训练图像，各个训练图像中携带标签信息，标签信息包括目标对象和对应的位置信息，输入各个训练图像至初始目标检测模型，输出各个训练图像的预测标签，预测标签包括预测目标对象和预测位置信息；根据各个训练图像的标签信息和对应的预测标签更新初始目标检测模型的模型参数，直至初始目标检测模型满足预设收敛条件，得到预设目标检测模型。
图6示出了一个实施例中计算机设备的内部结构图。该计算机设备具体可以是图1中的拍摄设备110(或计算机设备120)。如图6所示，该计算机设备通过系统总线连接的处理器、存储器、网络接口、输入装置和显示屏。其中，存储器包括非易失性存储介质和内存储器。该计算机设备的非易失性存储介质存储有操作系统，还可存储有计算机程序，该计算机程序被处理器执行时，可使得处理器实现喷头控制方法。该内存储器中也可储存有计算机程序，该计算机程序被处理器执行时，可使得处理器执行喷头控制方法。计算机设备的显示屏可以是液晶显示屏或者电子墨水显示屏，计算机设备的输入装置可以是显示屏上覆盖的触摸层，也可以是计算机设备外壳上设置的按键、轨迹球或触控板，还可以是外接的键盘、触控板或鼠标等。
本领域技术人员可以理解，图6中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
在一个实施例中，本申请提供的喷淋装置可以实现为一种计算机程序的形式，计算机程序可在如图6所示的计算机设备上运行。计算机设备的存储器中可存储组成该喷淋装置的各个程序模块，比如，图5所示的图像获取模块201、目标检测模块202、属性信息获取模块203匹配模块204和控制模块205。各个程序模块构成的计算机程序使得处理器执行本说明书中描述的本申请各个实施例的喷头控制装置中的步骤。
例如，图6所示的计算机设备可以通过如图5所示的喷淋装置中的图像获取模块201执行实时获取图像，图像中包含至少一个目标对 象。计算机设备可通过目标检测模块202执行输入图像至预设目标检测模型，输出各个目标对象和对应的位置信息。计算机设备可通过属性信息获取模块203执行获取喷头的属性信息，属性信息包括喷淋区域和喷淋状态。计算机设备可通过匹配模块204根据各个目标对象的位置信息和喷淋区域匹配目标喷头。计算机设备可通过控制模块205执行当目标喷头的喷淋状态为关闭状态时，生成用于开启目标喷头的指令，发送指令。
在一个实施例中，提供了一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，处理器执行计算机程序时实现以下步骤：实时获取图像，图像中包含至少一个目标对象；输入图像至预设目标检测模型，输出各个目标对象和对应的位置信息；获取喷头的属性信息，属性信息包括喷淋区域和喷淋状态；根据各个目标对象的位置信息和喷淋区域匹配目标喷头；当目标喷头的喷淋状态为关闭状态时，生成用于开启目标喷头的指令，发送指令。
在一个实施例中，根据目标对象的位置信息和喷淋区域匹配目标喷头，包括：根据目标对象的位置信息查找候选喷淋区域，得到至少一个候选喷淋区域；计算目标对象位于各个候选喷淋区域的面积占比；将面积占比大于预设占比对应的候选喷淋区域，作为目标喷淋区域；将目标喷淋区域对应的喷头作为目标喷头。
在一个实施例中，处理器执行计算机程序时还实现以下步骤：生成预设目标检测模型，包括：获取多个训练图像，各个训练图像中携带标签信息，标签信息包括目标对象和对应的位置信息；输入各个训练图像至初始目标检测模型，输出各个训练图像的预测标签，预测标签包括预测目标对象和预测位置信息；根据各个训练图像的标签信息和对应的预测标签更新初始目标检测模型的模型参数，直至初始目标 检测模型满足预设收敛条件，得到预设目标检测模型。
在一个实施例中，预设目标检测模型包括编码模块和解码模块，编码模块对应的网络结构与ENet网络中的前15个模块相同，解码模块包括反卷积单元和全卷积单元，全卷积单元的输入为反卷积单元的输出。
在一个实施例中，编码模块包括下采样单元和卷积单元，输入图像至预设目标检测模型，输出各个目标对象和对应的位置信息，包括：输入图像至下采样单元执行下次采样操作，得到图像的下采样特征；输入图像的下采样特征至卷积单元执行卷积操作，得到图像的卷积特征；输入图像的卷积特征至反卷积单元执行反卷积操作，得到各个图像的反卷积特征；输入图像的反卷积特征至全卷积单元执行全卷卷积操作，输出图像的目标对象和对应的位置信息。
在一个实施例中，反卷积单元的网络结构为第一反卷积子单元后接两个bottleneck结构，两个bottleneck结构后接第二反卷积子单元，第二反卷积子单元后接一个bottleneck结构。
在一个实施例中，提供了一种计算机可读存储介质，其上存储有计算机程序，计算机程序被处理器执行时实现以下步骤：实时获取图像，图像中包含至少一个目标对象；输入图像至预设目标检测模型，输出各个目标对象和对应的位置信息；获取喷头的属性信息，属性信息包括喷淋区域和喷淋状态；根据各个目标对象的位置信息和喷淋区域匹配目标喷头；当目标喷头的喷淋状态为关闭状态时，生成用于开启目标喷头的指令，发送指令。
在一个实施例中，根据目标对象的位置信息和喷淋区域匹配目标喷头，包括：根据目标对象的位置信息查找候选喷淋区域，得到至少一个候选喷淋区域；计算目标对象位于各个候选喷淋区域的面积占比； 将面积占比大于预设占比对应的候选喷淋区域，作为目标喷淋区域；将目标喷淋区域对应的喷头作为目标喷头。
在一个实施例中，计算机程序被处理器执行时还实现以下步骤：生成预设目标检测模型，包括：获取多个训练图像，各个训练图像中携带标签信息，标签信息包括目标对象和对应的位置信息；输入各个训练图像至初始目标检测模型，输出各个训练图像的预测标签，预测标签包括预测目标对象和预测位置信息；根据各个训练图像的标签信息和对应的预测标签更新初始目标检测模型的模型参数，直至初始目标检测模型满足预设收敛条件，得到预设目标检测模型。
在一个实施例中，预设目标检测模型包括编码模块和解码模块，编码模块对应的网络结构与ENet网络中的前15个模块相同，解码模块包括反卷积单元和全卷积单元，全卷积单元的输入为反卷积单元的输出。
在一个实施例中，编码模块包括下采样单元和卷积单元，输入图像至预设目标检测模型，输出各个目标对象和对应的位置信息，包括：输入图像至下采样单元执行下次采样操作，得到图像的下采样特征；输入图像的下采样特征至卷积单元执行卷积操作，得到图像的卷积特征；输入图像的卷积特征至反卷积单元执行反卷积操作，得到各个图像的反卷积特征；输入图像的反卷积特征至全卷积单元执行全卷卷积操作，输出图像的目标对象和对应的位置信息。
在一个实施例中，反卷积单元的网络结构为第一反卷积子单元后接两个bottleneck结构，两个bottleneck结构后接第二反卷积子单元，第二反卷积子单元后接一个bottleneck结构。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的程序可存储于一非易失性计算机可读取存储介质中，该程序在执行时， 可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
需要说明的是，在本文中，诸如“第一”和“第二”等之类的关系术语仅仅用来将一个实体或者操作与另一个实体或操作区分开来，而不一定要求或者暗示这些实体或操作之间存在任何这种实际的关系或者顺序。而且，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者设备不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括所述要素的过程、方法、物品或者设备中还存在另外的相同要素。
以上所述仅是本申请的具体实施方式，使本领域技术人员能够理解或实现本申请。对这些实施例的多种修改对本领域的技术人员来说将是显而易见的，本文中所定义的一般原理可以在不脱离本申请的精神或范围的情况下，在其它实施例中实现。因此，本申请将不会被限制于本文所示的这些实施例，而是要符合与本文所申请的原理和新颖特点相一致的最宽的范围。

Claims (14)

  1. 一种喷头控制方法，所述方法包括：
    实时获取图像，所述图像中包含至少一个目标对象；
    输入所述图像至预设目标检测模型，输出各个所述目标对象和对应的位置信息；
    获取喷头的属性信息，所述属性信息包括喷淋区域和喷淋状态；
    根据各个所述目标对象的位置信息和所述喷淋区域匹配目标喷头；
    当所述目标喷头的喷淋状态为关闭状态时，生成用于开启所述目标喷头的指令，发送所述指令。
  2. 根据权利要求1所述的方法，所述根据所述目标对象的位置信息和所述喷淋区域匹配目标喷头，包括：
    根据所述目标对象的位置信息查找候选喷淋区域，得到至少一个候选喷淋区域；
    计算所述目标对象位于各个所述候选喷淋区域的面积占比；
    将所述面积占比大于预设占比对应的所述候选喷淋区域，作为目标喷淋区域；
    将所述目标喷淋区域对应的喷头作为所述目标喷头。
  3. 根据权利要求1所述的方法，生成所述预设目标检测模型的步骤，包括：
    获取多个训练图像，各个所述训练图像中携带标签信息，所述标签信息包括目标对象和对应的位置信息；
    输入各个所述训练图像至初始目标检测模型，输出各个所述训练 图像的预测标签，所述预测标签包括预测目标对象和预测位置信息；
    根据各个所述训练图像的标签信息和对应的所述预测标签更新所述初始目标检测模型的模型参数，直至所述初始目标检测模型满足预设收敛条件，得到所述预设目标检测模型。
  4. 根据权利要求1至3中任一项所述的方法，所述预设目标检测模型包括编码模块和解码模块，所述编码模块对应的网络结构与ENet网络中的前15个模块相同，所述解码模块包括反卷积单元和全卷积单元，所述全卷积单元的输入为所述反卷积单元的输出。
  5. 根据权利要求4所述的方法，所述编码模块包括下采样单元和卷积单元，所述输入所述图像至预设目标检测模型，输出各个所述目标对象和对应的位置信息，包括：
    输入所述图像至所述下采样单元执行下次采样操作，得到所述图像的下采样特征；
    输入所述图像的下采样特征至所述卷积单元执行卷积操作，得到所述图像的卷积特征；
    输入所述图像的卷积特征至所述反卷积单元执行反卷积操作，得到各个所述图像的反卷积特征；
    输入所述图像的反卷积特征至所述全卷积单元执行全卷积操作，输出所述图像的目标对象和对应的位置信息。
  6. 根据权利要求4所述的方法，所述反卷积单元的网络结构为第一反卷积子单元后接两个bottleneck结构，两个所述bottleneck结构后接第二反卷积子单元，所述第二反卷积子单元后接一个bottleneck结构。
  7. 一种喷头控制装置，所述装置包括：
    图像获取模块，用于实时获取图像，所述图像中包含至少一个目标对象；
    目标检测模块，用于输入所述图像至预设目标检测模型，输出各个所述目标对象和对应的位置信息；
    属性信息获取模块，用于获取喷头的属性信息，所述属性信息包括喷淋区域和喷淋状态；
    匹配模块，用于根据各个所述目标对象的位置信息和所述喷淋区域匹配目标喷头；
    控制模块，用于当所述目标喷头的喷淋状态为关闭状态时，生成用于开启所述目标喷头的指令，发送所述指令。
  8. 根据权利要求7所述的装置，所述匹配模块具体用于根据所述目标对象的位置信息查找候选喷淋区域，得到至少一个候选喷淋区域，计算所述目标对象位于各个所述候选喷淋区域的面积占比，将所述面积占比大于预设占比对应的所述候选喷淋区域，作为目标喷淋区域，将所述目标喷淋区域对应的喷头作为所述目标喷头。
  9. 根据权利要求7所述的装置，还包括：模型生成模块，用于获取多个训练图像，各个训练图像中携带标签信息，标签信息包括目标对象和对应的位置信息，输入各个训练图像至初始目标检测模型，输出各个训练图像的预测标签，预测标签包括预测目标对象和预测位置信息；根据各个训练图像的标签信息和对应的预测标签更新初始目标检测模型的模型参数，直至初始目标检测模型满足预设收敛条件，得到预设目标检测模型。
  10. 根据权利要求7至9中任一项所述的装置，所述预设目标检测模型包括编码模块和解码模块，所述编码模块对应的网络结构与ENet网络中的前15个模块相同，所述解码模块包括反卷积单元和全卷 积单元，所述全卷积单元的输入为所述反卷积单元的输出。
  11. 根据权利要求10所述的装置，所述编码模块包括下采样单元和卷积单元，所述目标检测模块具体用于输入所述图像至所述下采样单元执行下次采样操作，得到所述图像的下采样特征；输入所述图像的下采样特征至所述卷积单元执行卷积操作，得到所述图像的卷积特征；输入所述图像的卷积特征至所述反卷积单元执行反卷积操作，得到各个所述图像的反卷积特征；输入所述图像的反卷积特征至所述全卷积单元执行全卷积操作，输出所述图像的目标对象和对应的位置信息。
  12. 根据权利要求10所述的装置，所述反卷积单元的网络结构为第一反卷积子单元后接两个bottleneck结构，两个所述bottleneck结构后接第二反卷积子单元，所述第二反卷积子单元后接一个bottleneck结构。
  13. 一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现权利要求1至6中任一项所述方法的步骤。
  14. 一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现权利要求1至6中任一项所述的方法的步骤。
PCT/CN2020/124470 2019-11-05 2020-10-28 喷头控制方法、装置、计算机设备和存储介质 WO2021088700A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911073264.9A CN111027380A (zh) 2019-11-05 2019-11-05 喷头控制方法、装置、计算机设备和存储介质
CN201911073264.9 2019-11-05

Publications (1)

Publication Number Publication Date
WO2021088700A1 true WO2021088700A1 (zh) 2021-05-14

Family

ID=70200816

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/124470 WO2021088700A1 (zh) 2019-11-05 2020-10-28 喷头控制方法、装置、计算机设备和存储介质

Country Status (2)

Country Link
CN (1) CN111027380A (zh)
WO (1) WO2021088700A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116273692A (zh) * 2023-03-14 2023-06-23 深圳徕科技术有限公司 分布式高速喷射阀联合控制方法、装置、设备及存储介质
CN117180952A (zh) * 2023-11-07 2023-12-08 湖南正明环保股份有限公司 多向气流料层循环半干法烟气脱硫系统及其方法

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111027380A (zh) * 2019-11-05 2020-04-17 北京海益同展信息科技有限公司 喷头控制方法、装置、计算机设备和存储介质
CN111539937A (zh) * 2020-04-24 2020-08-14 北京海益同展信息科技有限公司 一种物体指标的检测方法、家畜重量的检测方法及装置
CN112766171B (zh) * 2021-01-21 2023-01-31 京东科技信息技术有限公司 一种喷淋方法、装置、系统及介质
CN113312799B (zh) * 2021-06-25 2023-03-21 中铁十一局集团桥梁有限公司 喷淋系统冲洗方法、装置、设备及可读存储介质
CN114445027B (zh) * 2021-12-27 2023-04-07 江西省交通工程集团有限公司 用于喷淋车的出勤频率控制方法与系统

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103530643A (zh) * 2013-10-11 2014-01-22 中国科学院合肥物质科学研究院 基于农作物行间自动识别技术的农药定位喷洒方法及系统
KR20140105058A (ko) * 2013-02-21 2014-09-01 순천대학교 산학협력단 센서 네트워크 기반 차량 방역 시스템
CN107038448A (zh) * 2017-03-01 2017-08-11 中国科学院自动化研究所 目标检测模型构建方法
CN107950506A (zh) * 2017-11-15 2018-04-24 广州极飞科技有限公司 移动装置、基于移动装置的喷洒控制方法及装置
CN108734277A (zh) * 2018-05-18 2018-11-02 拓普暨达（广州）基因精准医疗科技有限公司 一种基于卷积神经网络的靶向药物喷洒方法
CN111027380A (zh) * 2019-11-05 2020-04-17 北京海益同展信息科技有限公司 喷头控制方法、装置、计算机设备和存储介质

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109409275B (zh) * 2018-10-18 2021-06-15 广州极飞科技股份有限公司 目标对象的识别方法及装置、施药信息的确定方法
CN109446987A (zh) * 2018-10-29 2019-03-08 北京麦飞科技有限公司 基于pca和pnn算法检测水稻病虫害等级的方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20140105058A (ko) * 2013-02-21 2014-09-01 순천대학교 산학협력단 센서 네트워크 기반 차량 방역 시스템
CN103530643A (zh) * 2013-10-11 2014-01-22 中国科学院合肥物质科学研究院 基于农作物行间自动识别技术的农药定位喷洒方法及系统
CN107038448A (zh) * 2017-03-01 2017-08-11 中国科学院自动化研究所 目标检测模型构建方法
CN107950506A (zh) * 2017-11-15 2018-04-24 广州极飞科技有限公司 移动装置、基于移动装置的喷洒控制方法及装置
CN108734277A (zh) * 2018-05-18 2018-11-02 拓普暨达（广州）基因精准医疗科技有限公司 一种基于卷积神经网络的靶向药物喷洒方法
CN111027380A (zh) * 2019-11-05 2020-04-17 北京海益同展信息科技有限公司 喷头控制方法、装置、计算机设备和存储介质

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116273692A (zh) * 2023-03-14 2023-06-23 深圳徕科技术有限公司 分布式高速喷射阀联合控制方法、装置、设备及存储介质
CN116273692B (zh) * 2023-03-14 2023-08-15 深圳徕科技术有限公司 分布式高速喷射阀联合控制方法、装置、设备及存储介质
CN117180952A (zh) * 2023-11-07 2023-12-08 湖南正明环保股份有限公司 多向气流料层循环半干法烟气脱硫系统及其方法
CN117180952B (zh) * 2023-11-07 2024-02-02 湖南正明环保股份有限公司 多向气流料层循环半干法烟气脱硫系统及其方法

Also Published As

Publication number Publication date
CN111027380A (zh) 2020-04-17

Similar Documents

Publication Publication Date Title
WO2021088700A1 (zh) 喷头控制方法、装置、计算机设备和存储介质
Zhang et al. Surface defect detection of steel strips based on classification priority YOLOv3-dense network
CN111079632A (zh) 文本检测模型的训练方法、装置、计算机设备和存储介质
CN109308488B (zh) 乳腺超声图像处理装置、方法、计算机设备及存储介质
JP2022532460A (ja) モデル訓練方法、装置、端末及びプログラム
Francies et al. A robust multiclass 3D object recognition based on modern YOLO deep learning algorithms
LU502731B1 (en) Method for monitoring abnormality of power production, apparatus, computer device, and storage medium therefor
WO2021180030A1 (zh) 行为识别方法及系统、电子设备和计算机可读存储介质
US20210342593A1 (en) Method and apparatus for detecting target in video, computing device, and storage medium
WO2023130648A1 (zh) 一种图像数据增强方法、装置、计算机设备和存储介质
WO2023206904A1 (zh) 一种行人轨迹跟踪方法、系统及相关装置
JP2023162341A (ja) 拡張データによって機械モデルを訓練するためのシステムおよび方法
Liu et al. Lightweight object detection algorithm for robots with improved YOLOv5
CA3193958A1 (en) Processing images using self-attention based neural networks
CN111429424B (zh) 一种基于深度学习的加热炉入口异常识别方法
CN110751056B (zh) 基于改进自顶向下法多人姿态检测的行人运动预测方法
CN111832561B (zh) 基于计算机视觉的字符序列识别方法、装置、设备和介质
CN113706481A (zh) 精子质量检测方法、装置、计算机设备和存储介质
CN111199535A (zh) 一种动物状态监测方法、装置、电子设备及存储介质
CN111428664A (zh) 一种基于人工智能深度学习技术的计算机视觉的实时多人姿态估计方法
WO2023169053A1 (zh) 基于相机阵列的目标追踪方法及系统
CN111738074B (zh) 基于弱监督学习的行人属性识别方法、系统及装置
Zeng et al. Nlfftnet: A non-local feature fusion transformer network for multi-scale object detection
CN111080697B (zh) 检测目标对象方向的方法、装置、计算机设备和存储介质
CN115797781A (zh) 农作物识别方法、装置、计算机设备和存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20884700

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20884700

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 02/09/2022)

122 Ep: pct application non-entry in european phase

Ref document number: 20884700

Country of ref document: EP

Kind code of ref document: A1