WO2021208407A1 - 目标物检测方法、装置和图像采集方法、装置 - Google Patents

目标物检测方法、装置和图像采集方法、装置 Download PDF

Info

Publication number
WO2021208407A1
WO2021208407A1 PCT/CN2020/125251 CN2020125251W WO2021208407A1 WO 2021208407 A1 WO2021208407 A1 WO 2021208407A1 CN 2020125251 W CN2020125251 W CN 2020125251W WO 2021208407 A1 WO2021208407 A1 WO 2021208407A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
plants
target
sampling
sampled
Prior art date
Application number
PCT/CN2020/125251
Other languages
English (en)
French (fr)
Inventor
陈洪生
董雪松
Original Assignee
苏州极目机器人科技有限公司
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 苏州极目机器人科技有限公司 filed Critical 苏州极目机器人科技有限公司
Priority to BR112022020889A priority Critical patent/BR112022020889A2/pt
Publication of WO2021208407A1 publication Critical patent/WO2021208407A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/20—Scenes; Scene-specific elements in augmented reality scenes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06M—COUNTING MECHANISMS; COUNTING OF OBJECTS NOT OTHERWISE PROVIDED FOR
    • G06M11/00—Counting of objects distributed at random, e.g. on a surface
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/07—Target detection

Definitions

  • This application relates to the technical field of image detection applications, for example, to a target detection method and device, and an image acquisition method and device.
  • the method of emasculation detection by seed production companies is generally: walking to multiple sampling points manually, sampling multiple plants in the area, and checking whether tassel removal has been completed.
  • the detection effect is poor and it takes multiple days to repeat the sampling, otherwise the emasculation rate cannot be guaranteed.
  • This method wastes time and increases labor costs. Due to time-consuming, labor-intensive and high cost, it is difficult to sample a large number of samples to ensure the purity of emasculation.
  • This application provides a target detection method, device, and image acquisition method and device, which collect highly consistent sampling images through machine control, perform target recognition and removal degree statistics, greatly improve the detection efficiency of target removal, and greatly shorten the target
  • the detection time of object removal greatly reduces the cost of target detection, and completely avoids all risks caused by the need to go deep into the field when the target removal detection personnel must work in the field.
  • the embodiment of the present application provides a target detection method, including:
  • the target objects of all maternal line plants in the sampled image are identified to perform target removal detection.
  • An embodiment of the present application provides a target detection device, including:
  • the sampling image acquisition module is configured to acquire sampling images of sampling points in a regularly planted plant area.
  • the plant area includes male and female line plants.
  • the sampling image is collected from directly above the plant area.
  • the sampled image includes all the female line plants except the male line plants;
  • the removal detection module is configured to identify the target objects of all maternal line plants in the sampled image according to the sampled image to perform target removal detection.
  • An embodiment of the present application provides an image acquisition method, including:
  • the target sampling posture of the image acquisition device is determined according to the image information, the target sampling posture includes at least one of the following: a collection height and a collection angle, where the collection angle is the horizontal axis of the image of the image collection device and the planting The angle of the extension direction of the row;
  • the embodiment of the present application also provides an image acquisition device, including:
  • An image acquisition module configured to acquire image information of sampling points in a regularly planted plant area, where the plant area includes male line plants and female line plants;
  • the posture determination module is configured to determine the target sampling posture of the image acquisition device according to the image information, and the target sampling posture includes at least one of the following: a collection height and a collection angle, wherein the collection angle is the size of the image collection device The angle between the horizontal axis of the image and the extension direction of the planting row;
  • the image determining module is configured to obtain the sampled image corresponding to the sampling point based on the target sampling posture.
  • An embodiment of the present application also provides an electronic 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 to implement the above-mentioned target detection method or image acquisition method.
  • the embodiment of the present application also provides a computer-readable storage medium, and a computer program is stored on the computer-readable storage medium, and the computer program executes the above-mentioned target detection method or image acquisition method when the computer program is run by a processor.
  • FIG. 1 is a schematic diagram of the planting of a male parent plant and a female parent plant provided by an embodiment of the application;
  • FIG. 2 is a flowchart of an image acquisition method provided by an embodiment of the application
  • FIG. 3 is a schematic diagram of image information of an image acquisition device in a ready position according to an embodiment of the application
  • FIG. 4 is a schematic diagram of image information of an image acquisition device in a sampling position provided by an embodiment of the application;
  • FIG. 5 is a flowchart of a target detection method provided by an embodiment of the application.
  • FIG. 6 is a schematic diagram of a user interface for emasculation detection provided by an embodiment of the application.
  • FIG. 7 is a schematic diagram of an application scenario of a emasculation detection method provided by an embodiment of the application.
  • FIG. 8 is a functional module diagram of an image acquisition device provided by an embodiment of the application.
  • FIG. 9 is a functional block diagram of a target detection device provided by an embodiment of the application.
  • FIG. 10 is a schematic diagram of the hardware architecture of an electronic device according to an embodiment of the application.
  • Removal detection is widely used in the field of agricultural breeding, taking maize emasculation as an example.
  • the general planting area is more than 1,000 acres, and the planting area plot boundary is more than 1 km.
  • the plants are planted in rows.
  • the female parent row is first removed by mechanical emasculation.
  • the male tassels are reserved for pollination.
  • inter-row planting is adopted, that is, the male and female parent rows are planted at intervals, as shown in Figure 1.
  • the tassels of the female parent row are removed, leaving only the tassels of the male parent row.
  • the fruit on the female parent row plants is a combination of the pollen of the male parent and the eggs of the female parent. Realize hybrid seed production. If the female parent is not well emasculated, the pollen of the female parent will be pollinated on its own ears to form selfed seeds, which will greatly affect the purity of the seeds.
  • the planting rules of plants can be shown in Figure 1, for example: female parent row 4 rows, male parent row 2 rows, female parent row 4 rows, male parent row 2 rows, and the row spacing is determined; it can also be female parent row 6 Rows, 2 rows from the male parent, 6 rows from the female parent, 2 rows from the male parent, planted at intervals. This is just an example. There is no restriction on the number of parent rows and the number of parent rows, and there is no restriction on the line spacing.
  • the detection process is complex and manual participation is high, making it difficult to effectively improve the efficiency and accuracy of emasculation detection.
  • the target detection method, device, and image acquisition method and device provided by the embodiments of the application collect highly consistent sampling images through a machine control method to ensure the efficiency and accuracy of target removal detection, save time and effort, and reduce The cost of testing.
  • an image acquisition method disclosed in the embodiment of the application is first introduced, which is mainly applied to control equipment, such as aircraft, in a plant scene adapted to regular planting.
  • regular planting includes but is not limited to Planting along the line or in a community, such as regularly planted corn, rice, soybeans, rape, etc.
  • the embodiment of the present application takes corn emasculation as an example for illustration.
  • Fig. 2 is a flowchart of an image acquisition method provided by an embodiment of the application.
  • an image acquisition method provided by the present application mainly includes the following steps:
  • Step S102 Obtain image information of sampling points in a regularly planted plant area.
  • the plant area includes male row plants and female row plants.
  • Step S104 Determine the target sampling posture of the image collection device according to the image information, the target sampling posture includes at least one or more of the following: collection height and collection angle, the collection angle being the extension of the image horizontal axis of the image collection device and the planting row The angle of the direction.
  • Step S106 Acquire a sampled image corresponding to the sample point based on the target sample posture.
  • the acquisition height of the image acquisition device is determined according to the acquired image information of the sampling points of the plant area, and the acquisition angle reaches the target sampling attitude, and the sampling images with the same height are collected at the sampling points according to the target sampling attitude, Target removal detection is performed according to the plant conditions in the sampled image, such as emasculation detection.
  • the embodiment of the application collects highly consistent sampling images through a machine control method to accurately identify the number of missing tassels and plants, thereby improving the emasculation detection.
  • the degree of automation while ensuring the accuracy of emasculation detection, saves time and effort.
  • the embodiment of the present application takes emasculation detection as an example for description, and is not limited to this, and is also applicable to other target removal detection scenarios.
  • Standardized acquisition of highly consistent sampling images through machine control methods mainly for regularly planted plant scenes (higher consistency of planting rules), through the image acquisition method of this application (highly consistent acquisition methods), a high degree of consistency can be obtained
  • Sampling images to ensure that the actual area of each sampled image is the same
  • the number of plants sampled is highly consistent (generally, the number of plants in each sampled image is the same), so that sampling inspection can be performed well.
  • the remote control device can be used to manually determine the plane position and height position of the aircraft, so as to ensure the consistency of the captured images and eliminate other plant rows. Interference to improve the consistency of the sampled image.
  • the boundary between the paternal line and the maternal line is confirmed by human eyes, and the collection height and angle are determined when the interference of the paternal line is eliminated during sampling, which ensures the accuracy and avoids the difference between the paternal line and the maternal line.
  • the appearance is very similar, and the machine recognizes the difficult points that are difficult to accurately distinguish.
  • the number of plants displayed in a sampled image can be obtained (either by manual counting or image recognition, without limitation here), and then the total number of plants in all the sampled images can be obtained for subsequent statistics.
  • the tassel image of the female parent line is collected by the image acquisition equipment to detect whether the emasculation meets the requirements, which improves the detection efficiency and saves manpower and material resources.
  • the tassel image is collected by the remote control aircraft for tassel sampling, avoiding manual operations, avoiding the risk of going deep into the farmland, saving manpower and material resources, and improving the detection effect and efficiency.
  • the attitude of the aircraft can be controlled by the host computer, and then the aircraft is equipped with an image acquisition device (camera) for image acquisition.
  • an image acquisition device camera
  • step S104 further includes the following steps:
  • Step 1.1 adjust the current height of the image acquisition device until the image information includes all the female line plants except the male line plants, and determine the acquisition height in the target sampling posture.
  • the aerial height at this time is the acquisition height in the target sampling attitude.
  • the shooting boundary is based on the male parent plant. In the application scene of other continuous planting rows, it can be set as needed.
  • the target sampling posture includes the same collection direction, and the collection direction is the direction in which the image collection device faces the ground, such as vertical downwards.
  • the aerial lens is required to shoot the plant vertically downwards to obtain sampled images to identify the top of the tassel and improve the recognition rate of the tassel. If it is tilted, part of the body of the tassel will be recognized , The overlapping tassels will affect the recognition accuracy.
  • step S104 can also be implemented through the following steps:
  • Step 1.2 adjust the current acquisition angle of the image acquisition device until the horizontal axis of the image of the image acquisition device and the extension direction of the planting row in the image information are at a preset angle, and determine the acquisition angle in the target sampling posture.
  • the current acquisition angle of the image acquisition device is adjusted until the horizontal axis of the image of the image acquisition device is parallel or perpendicular to the extension direction of the planting row in the image information, the acquisition angle in the target sampling posture is determined, and the acquisition The angle is 0° or 90°. That is, the extension direction of the planting row in the collected image is parallel or perpendicular to the image boundary line, and the collection angle of the image collection device can be selected as 0° or 90°.
  • the method provided in the embodiment of the present application further includes:
  • Step 1.3 controlling the exposure of the sampled image according to the brightness of the image information.
  • the brightness of the aerial photography is controlled. In turn, the quality of the sampled image is ensured, and the situation where the sampled image cannot be recognized is prevented.
  • the plant area includes a plurality of sampling points, and the sampling images corresponding to each sampling point do not overlap with each other.
  • the general plant area can include multiple sampling points. It is necessary to ensure that the sampling images collected by multiple sampling points do not overlap with each other to avoid statistical errors in the number of tassels and the number of plants, thereby ensuring emasculation detection The accuracy of the results.
  • first control aircraft remote control or host computer control
  • aerial photography aircraft to fly above the target sampling point to be collected, hover and adjust the flying height of the aircraft, the lens acquisition angle and acquisition direction meet the above requirements , And ensure that the target parent bank and the parent bank on both sides of the target parent bank are all located in the screen, as shown in Figure 3 for the preparation position of the aerial aircraft.
  • the aerial vehicle position as shown in Figure 4 adjusts the exposure to the brightest part of the picture without white spots, and then take the image capture. If the sampled image contains only part of the maternal line, for example, the sampled image includes 3 lines of maternal lines, or 2 lines of maternal lines, it will affect the accuracy of sampling.
  • step S102 further includes the following steps:
  • Step 2.1 obtain the sampling points in the regularly planted plant area and the position coordinates of the sampling points;
  • Step 2.2 according to the location coordinates of the sampling point, plan the flight path from the image acquisition device to the sampling point;
  • Step 2.3 obtain image information based on the flight path.
  • the execution subject of the method of acquiring images is the flight controller (controlling the aircraft equipped with image acquisition equipment) or the aircraft equipped with image acquisition equipment. This is achieved by controlling the aircraft to reach the sampling point and controlling the image acquisition equipment to collect the sampled images.
  • the aircraft receives the position coordinates of the sampling point in the planting area, and the aircraft arrives above the sampling point according to the flight path; the sampling point may be random or preset.
  • the aircraft can reach multiple sampling points remotely by the operator, or the aircraft can plan the flight path according to the preset or randomly set multiple sampling point positions in the planting area, and automatically reach the multiple sampling points according to the planned path for image collection.
  • the image information collected based on the foregoing method includes the position coordinates of the sampling point.
  • the sampling image includes position information, such as position information corresponding to the sampling point. Based on the location information corresponding to the sampling point, in order to locate the accurate location of the target.
  • the image acquisition device is controlled to collect the image information of the sampling point, obtain the sampled image, and save it. Repeat the previous steps until the number of sampled images obtained meets the requirements. In order to obtain the results of emasculation detection more accurately, a large number of sampled images need to be collected.
  • the above image sampling step needs to be repeated in the above-mentioned planting area. This step needs to be repeated multiple times to ensure continuous emasculation detection during plant growth and prevent male growth afterwards. Spike was not detected.
  • the sampling location Before determining the sampling location, it is also necessary to confirm whether the photographed images include all the maternal rows of continuous planting rows (machine judgment or manual judgment). For example, if the number of rows of the maternal row for continuous planting is known to be 4 rows, It is judged whether the photographed image contains a four-line maternal line. If only three lines are included, one line is omitted and the sampling is not performed. This situation will affect the correctness of the sampling results.
  • the number of rows is known and can be preset according to the planting rules.
  • the sampled image is a bottom view, which is an image with a vertical downward perspective of the lens, and the posture of the lens of the image acquisition device is fixed (the same) when different images are acquired.
  • the axis of the lens of the image capture device can be fixed to the vertical direction.
  • the operator can remotely control the aircraft to reach a random or set sampling point in the plant area, adjust the collection direction, and reduce the flying height of the remote control aircraft. While the height is lowered, the operator can use the image acquisition device Real-time image captured, confirm whether the captured image only includes the maternal line, and the extension direction of the corn line is parallel or perpendicular to the horizontal axis of the screen, if the conditions are met, and include all continuously planted female lines, control the image acquisition equipment to collect Image; if the captured image also includes the paternal line or contains only a small number of consecutive female lines, or the extension direction of the planting line is not parallel or perpendicular to the horizontal axis of the screen, adjust the position of the aircraft while reducing the height until only the captured image After the maternal line is included and all consecutive maternal lines are included, and the extension direction of the planting line is parallel or perpendicular to the horizontal axis of the image, the image at that position can be collected.
  • the flying height of the aircraft is not set in advance, but is dynamically determined during shooting based on the row spacing of plants, the number of rows of planting parent rows, the column spacing of plants, and the size of the aerial image capture device's field of view.
  • the image acquisition device can obtain 80 pictures at the sampling point, and each picture can contain 30 plants, so one sampling point can obtain the emasculation information of 2400 plants, which greatly improves the emasculation detection. Efficiency, the number of pictures and the number of plants are not limited, only examples.
  • the operator remotely controls the aircraft to reach another random or set sampling point, and repeats the above operations until the collection of multiple sampling points is completed.
  • the number of sampling points can be 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, etc., without limitation.
  • the location information of the planting area is obtained, as well as the row spacing of the plants, the column spacing of the plants, the number of rows of planting female parent rows, and the number of rows of planting male parent rows, etc.
  • you can plan The flight path of the image collected by the aircraft can be automatically collected.
  • the target detection method can be executed to achieve the purpose of removing the target, as shown in FIG. 5, including the following step:
  • Step S202 Obtain sampling images of sampling points in the regularly planted plant area.
  • the plant area includes male row plants and female row plants.
  • the sampling image is collected from directly above the plant area.
  • the sampled image includes all but the male row plants.
  • step S204 the target objects of all female parent plants in the sampled image are identified according to the sampled image to perform target removal detection.
  • the sampling posture of the image acquisition device is dynamically adjusted to collect the sampling images that meet the requirements, the historical sampling images are marked and the deep learning model is trained to accurately identify the target, and the successful deep learning is used
  • the model checks the sampled images to obtain the number and proportion of the target, so as to determine the degree of removal of the target in the plant area, and thereby ensure the purity of the seed.
  • This application uses machines instead of manuals to perform automatic sample collection, target identification and removal degree statistics, greatly improve target removal detection efficiency, greatly shorten target removal detection time, greatly reduce target detection costs, and completely avoid target removal Inspectors must go deep into all the risks brought by field work.
  • the target identification is performed by collecting sampling images of all female row plants except the male row plants collected directly above the plant area, thereby achieving the target removal situation. Detection.
  • the horizontal axis of the sampled image and the extension direction of the planting row are at a preset angle. In order to improve the accuracy of target detection and prevent missed detection.
  • the plant area includes multiple sampling points, and sampling images corresponding to the multiple sampling points do not overlap each other. Further improve the efficiency and accuracy of target identification and detection.
  • the sampled image includes position information for subsequent positioning of the target object.
  • the target includes any one or both of tassels and flower buds.
  • tassels and flower buds can be detected. The difficulty of detecting flower buds is far greater than that of tassels, so it can be used as a reference detection scheme.
  • the sampled images satisfying the foregoing embodiments have higher consistency, and in the image recognition process, the removal of the target object can be detected more accurately, and more accurate detection results can be obtained.
  • the aircraft image acquisition device collects the sampled image and stores it in the memory so that after the sampling is completed, it can import a preset recognition model to perform recognition processing and reduce the cost of the aircraft.
  • computing and processing equipment can also be installed on the aircraft. The sampled images are calculated and processed in real time during flight, and the recognition results can be obtained quickly.
  • the method provided in the embodiment of the application collects sampled images by specifying the height, angle and direction of the target object collection, which ensures the consistency of the obtained images, thereby improving the recognition accuracy; at the same time, the parent line is excluded by controlling the preset flying height.
  • the interference can ensure the identification of all consecutively planted female lines, avoid the omission of the female lines, and further improve the accuracy of identification and detection.
  • step S204 can also be implemented by the following steps:
  • Step 3.1 training a deep learning model based on historical sampled images to obtain a target object model.
  • a training set of labeled sampled images can be formed by labeling historical sampled images, so as to train a deep learning model, and then obtain a target object model.
  • Step 3.2 identify the target objects of all female parent plants in the sampled image.
  • the target objects of all the female parent plants in the unlabeled sampling image can be identified according to the target object model.
  • the deep learning model includes any open source or self-developed neural network for target detection based on deep learning. Sampling images based on the same standard are used to train deep learning models and identify detection targets, further improving the accuracy of target detection.
  • step S204 further includes:
  • Step 4.1 determine the number of plants of all female parent plants in the sampled image
  • the number of plants in the multiple sampled images obtained are respectively c1, c2, c3....cn.
  • sampling points can be randomly selected, and the sampling points can be determined in advance according to the planting area, and there is no restriction.
  • the number of sampling points is not limited. In order to improve the detection accuracy, the number of sampling points can be appropriately increased. The number of sampling points can also be determined according to the resolution of the image acquisition device, flying height, planting row spacing, planting column spacing, etc., to improve sampling accuracy.
  • the sampled image can be stored in the aircraft's memory, or it can be sent back in real time without limitation.
  • the sampled image only includes the maternal row plants and excludes the paternal row plants, which improves the recognition accuracy and avoids the interference of the paternal row.
  • the sampled image includes all the maternal line plants of successive planting rows to avoid omissions and prevent missed inspections.
  • the number of plants included in the sampled image is greater than or equal to the preset value, to prevent the number of plants from being too small and the detection efficiency is low, and the number of plants is also less than the second preset value, to avoid low resolution and inaccurate detection, and to ensure sampling statistics accuracy.
  • the image can be provided by image acquisition equipment such as surveying and mapping equipment, cameras, etc.
  • the image includes one or more of surveying and mapping image information and picture information, and is not limited to this.
  • Step 4.2 count the number of targets based on the identified targets.
  • the tassel texture feature in the sampled image can be recognized, and the number of tassels in the sampled image can be obtained; a large amount of historical data (training image) can be used to train the preset recognition model to obtain the deep learning model of the tassel.
  • the deep learning model uses training images to identify tassels in sampled images.
  • the method of collecting training images here is the same as the method of collecting sampled images in the previous embodiments. I will not repeat them here.
  • the training images and sampled images are consistent. Improve detection accuracy.
  • the sampled image is input into the deep learning model, and the deep learning model detects the sampled image to obtain tassel information. Extract the graphic characteristics of the tassel to be tested from the sampled images, and use the deep learning model to process the graphic characteristics of the tassel to be tested to obtain tassel information.
  • image processing can be performed on a special computing processing device; or each time the sampled image is acquired, it can be processed in real time, depending on whether the aircraft is equipped with a computing processor and computing processor. Computing power.
  • Step 4.3 determine the degree of target removal according to the number of plants and the number of targets.
  • the emasculation rate can be determined according to the total number of plants and the number of tassels; and then determine whether the emasculation rate reaches the emasculation rate threshold; if the emasculation rate reaches the emasculation rate threshold, the emasculation rate is qualified; if the emasculation rate does not reach the emasculation rate threshold , The castration is unqualified.
  • the preset threshold is the proportion of tassels that ensures that the seed meets the purity requirements. When d is less than or equal to the preset threshold, it means that the requirements for emasculation are met, and there is no need to emasculate again. When d is greater than or equal to the preset threshold, return Need to go to the cock again.
  • sampling image acquisition steps After a preset time interval, repeat the aforementioned sampling image acquisition steps in the same planting area. At this time, a sampling image of the same planting area is obtained.
  • the sampling image reflects the situation of the tassels after the plant has grown for a period of time. Or re-collect other sampling points in the same planting area to avoid undetected tassels, and on this basis, repeat multiple times to ensure that all growing tassels can be detected during a period of growth. , All meet the requirements of emasculation.
  • the sampled images collected in this application have the advantages of high resolution and good consistency.
  • the high resolution can maintain the texture characteristics of the tassels, and the high consistency makes the sample image characteristics of the training set and the detection set closer.
  • mature deep learning target detection technology is used, based on the training of a large number of historical sample images, It is easy to extract the characteristics of the sampled image, identify the tassels, and the computer automatically obtains the number of tassels in each sampled image.
  • the interference of the male parent on tassel counting is completely avoided.
  • the emasculation purity estimation method in the examples of this application is based on the above plant counting method and the high-precision corn tassel identification and counting method, and the estimated emasculation purity of the tested plant area can be obtained by the following formula:
  • Removal rate total number of targets in all sampled images detected / (total number of sampled images * average number of plants in each sampled image)
  • the removal rate can be calculated separately by time (days) according to the plot to find the trend of change.
  • the above-mentioned female parent row plant image is an image of a random area (sampling point) in the planting area or an image of a preset area (sampling point). No restrictions. Generally speaking, the more the number, the more accurate the removal rate obtained.
  • step 4.1) may also include the following steps:
  • Step 4.1.1 count the number of plants of all the female line plants in the sampled image
  • Step 4.1.2 identify all the female line plants in the sampled image, and obtain the number of plants
  • Step 4.1.3 obtain the number of plants of all female row plants in the preset number of sampled images, calculate the average number of plants of all the parent row plants in the preset number of sampled images, according to the average number of plants and the sampled image The total number of plants determines the number of plants in all maternal rows in the sampled image.
  • the total number of plants in the maternal row plant image collected in the planting area can be obtained, including only the maternal row and the maternal rows of all consecutive planting rows.
  • the number of plants in each sampled image is basically the same, so that each sampled image can be manually counted in advance from multiple sampled images.
  • the image-plant number mapping relationship generally varies according to the field and plant types. For a field, the plant types generally only need to be counted once, thereby avoiding the high cost of manual image recognition and counting accuracy problems caused by machine image recognition and counting; the number of plants can also be obtained through automatic machine recognition of sampled images; Or manually count each sampled image and count the number of plants.
  • each plot or set of plots with the same planting method and density randomly select multiple sampled images, manually count the number of plants contained in the multiple sampled images, and calculate the average of each sample The number of plants contained in the image C.
  • Target detection is not limited to one type, and multiple different targets can be detected on the same plot at the same time to obtain a thorough removal result.
  • the target detection method provided by the embodiment of the application uses a preset deep learning model to process the sampled image to obtain target information at the sampling point, which simplifies the statistical process of the target, thereby improving the statistical efficiency of the target, and avoids manual statistics.
  • the operator can implement emasculation detection through the user interface.
  • the user interface of the emasculation detection system includes user login and management system, picture upload and management system, and artificial intelligence (Artificial Intelligence). , AI) tassel identification system, identification result presentation and report system, user feedback system, area and billing system, AI tassel identification model, emasculation detection database and emasculation detection picture file system, which can realize the detection of males in sampled images Ears and plants are identified, and then the removal rate is calculated.
  • AI Artificial Intelligence
  • the removal rate after calculating the removal rate, it is also possible to automatically detect whether there is a target in the planting area by random inspection to determine whether the removal requirement is met.
  • the embodiment of this application implements target removal detection based on deep learning and aerial images, focusing on solving the following problems: 1. The problem of paternal and maternal identification; 2. Recognition accuracy; 3. Target proportion; 4. Operation Efficiency issues.
  • the aerial flying height, shooting angle, shooting direction, and sensitivity of the aircraft are controlled to obtain standardized and highly consistent aerial sampling pictures, realizing the number of plants and reducing the recognition of sampled images Difficulty, improve detection accuracy, and achieve accurate and efficient target removal detection results.
  • the embodiment further provides an image acquisition device 800, including:
  • the image acquisition module 801 is configured to acquire the image information of the sampling points of the regularly planted plant area, the plant area includes the male line plants and the female line plants;
  • the posture determination module 802 is configured to determine the target sampling posture of the image acquisition device according to the image information.
  • the target sampling posture includes at least one of the following: a collection height and a collection angle. The angle of the extension direction;
  • the image determination module 803 is configured to obtain a sampling image corresponding to the sampling point based on the target sampling posture, wherein the target removal detection is performed on the plants in the sampling image.
  • the sampled image includes location information.
  • the posture determination module 802 is configured to adjust the current height of the image capture device until the image information includes all female line plants except the male line plants, and determine the collection height in the target sampling posture.
  • the target collection posture includes a collection direction
  • the collection direction is a direction of the image collection device facing the ground, where the collection direction is vertical downward.
  • the posture determination module 802 is configured to adjust the current acquisition angle of the image acquisition device until the horizontal axis of the image of the image acquisition device is a preset angle with the extension direction of the planting row in the image information, and determine The acquisition angle in the target sampling attitude.
  • the image acquisition module 801 is configured to acquire sampling points and position coordinates of the sampling points in the regularly planted plant area; plan the flight path from the image acquisition device to the sampling points according to the position coordinates of the sampling points; Obtain image information based on the flight path.
  • the image acquisition device further includes controlling the exposure of the sampled image according to the brightness of the image information.
  • the embodiment further provides a target detection device 900, which includes:
  • the sampling image acquisition module 901 is configured to acquire sampling images of sampling points in a regularly planted plant area, where the plant area includes male and female line plants, and the sampling image is collected from directly above the plant.
  • the sampled image includes all the female line plants except the male line plants;
  • the removal detection module 902 is configured to identify the target objects of all maternal line plants in the sampled image according to the sampled image to perform target removal detection.
  • the horizontal axis of the sampled image and the extension direction of the planting row are at a preset angle.
  • the plant area includes multiple sampling points, and sampling images corresponding to the multiple sampling points do not overlap each other.
  • the sampled image includes location information.
  • the target includes at least one of a tassel and a flower bud.
  • the removal detection module 902 is configured to train a deep learning model according to the sampled image to obtain a target object model; and identify all maternal line plants in the sampled image according to the target object model. Target.
  • the deep learning model includes any open source or self-developed neural network based on deep learning target detection.
  • the removal detection module 902 is configured to train a deep learning model according to the sampled image to obtain a target object model by labeling the sampled image to form a training set of the labeled sampled image, according to The training set trains a deep learning model to obtain a target object model; the removal detection module 902 is configured to identify the target objects of all maternal plants in the sampled image according to the target object model in the following manner: The target object model recognizes the target objects of all the female parent plants in the unlabeled sampling images.
  • the removal detection module 902 is further configured to determine the number of plants of all maternal line plants in the sampled image; count the number of target objects according to the identified target; according to the number of plants and the target The number of objects determines the degree of target removal.
  • the removal detection module 902 is configured to determine the number of plants of all maternal line plants in the sampled image in the following manner: count the number of plants of all maternal line plants in the sampled image; Or, identify all maternal row plants in the sampled image to obtain the number of plants; or, obtain the number of plants of all maternal row plants in a preset number of sampled images, and calculate all maternal row plants in the preset number of sampled images.
  • the average number of plants in this row is determined based on the average number of plants and the total number of sampled images to determine the number of plants in all female rows in the sampled image.
  • FIG. 10 is a schematic diagram of the hardware architecture of an electronic device 1000 provided by an embodiment of the application.
  • the electronic device includes: a machine-readable storage medium 1001 and a processor 1002, and may also include a non-volatile storage medium 1003, a communication interface 1004, and a bus 1005; a machine-readable storage medium 1001, a processor 1002 , The non-volatile storage medium 1003 and the communication interface 1004 communicate with each other through the bus 1005.
  • the processor 1002 reads and executes the machine-executable instructions for target detection in the machine-readable storage medium 1001 to execute the target detection method or the image acquisition method described in the above embodiments.
  • the machine-readable storage medium mentioned herein can be any electronic, magnetic, optical or other physical storage device, and can contain or store information, such as executable instructions, data, and so on.
  • the machine-readable storage medium may be: Random Access Memory (RAM), volatile memory, non-volatile memory, flash memory, storage drive (such as hard drive), any type of storage disk (such as optical disk) , Digital Versatile Disc (DVD), etc.), or similar storage media, or a combination of them.
  • RAM Random Access Memory
  • volatile memory volatile memory
  • non-volatile memory flash memory
  • storage drive such as hard drive
  • any type of storage disk such as optical disk
  • DVD Digital Versatile Disc
  • similar storage media or a combination of them.
  • the non-volatile medium may be a non-volatile memory, flash memory, a storage drive (such as a hard disk drive), any type of storage disk (such as an optical disk, a DVD, etc.), or a similar non-volatile storage medium, or a combination thereof.
  • the computer-readable storage medium provided by the embodiment of the present application has a computer program stored in the readable storage medium, and the computer program code can realize the target detection method or the image acquisition method described in any of the above embodiments when the computer program code is executed
  • the computer program code can realize the target detection method or the image acquisition method described in any of the above embodiments when the computer program code is executed

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

Abstract

一种目标物检测方法、装置和图像采集方法、装置，目标物检测方法包括：获取规则种植的植株区域的采样点的采样图像（S202），所述植株区域包括父本行植株和母本行植株，所述采样图像从所述植株区域的正上方采集，所述采样图像包括除父本行植株外的全部母本行植株；根据所述采样图像识别所述采样图像中的全部母本行植株的目标物以进行目标物去除检测（S204）。

Description

目标物检测方法、装置和图像采集方法、装置
本申请要求在2020年04月16日提交中国专利局、申请号为202010302566.5的中国专利申请的优先权，该申请的全部内容通过引用结合在本申请中。
技术领域
本申请涉及图像检测应用的技术领域，例如涉及一种目标物检测方法、装置和图像采集方法、装置。
背景技术
制种公司在培育种子时，对种子纯度的要求达99.7％以上。对于玉米作物来说，为了获取高纯度种子，制种公司需要彻底去除玉米母本的雄穗，采用机械去雄或人工去雄的方法，并且在去雄的同时，及时检测去雄效果，确保去雄率，并在纯度不满足要求时重新去雄。
制种公司的去雄检测的方法一般是：人工行走至多个采样点，抽检区域内的多株植株，检查是否已经完成去除雄穗。检测效果较差且需要多天重复抽检，否则无法保证去雄率。这种方法既浪费时间，又增加人力成本。由于费时费力，成本高，难以大量取样保证去雄纯度。
发明内容
本申请提供一种目标物检测方法、装置和图像采集方法、装置，通过机器控制方式采集高度一致性的采样图像，进行目标物识别和去除程度统计，大幅提高目标物去除检测效率，大幅缩短目标物去除检测时间，大幅降低目标物检测成本，并彻底避免由于目标物去除检测人员必需深入田间工作时所带来的一切风险。
本申请的实施例提供一种目标物检测方法，包括：
获取规则种植的植株区域的采样点的采样图像，所述植株区域包括父本行植株和母本行植株，所述采样图像从所述植株区域的正上方采集，所述采样图像包括除父本行植株外的全部母本行植株；
根据所述采样图像识别所述采样图像中的全部母本行植株的目标物以进行 目标物去除检测。
本申请的实施例提供一种目标物检测装置，包括：
采样图像获取模块，设置为获取规则种植的植株区域的采样点的采样图像，所述植株区域包括父本行植株和母本行植株，所述采样图像从所述植株区域的正上方采集，所述采样图像包括除父本行植株外的全部母本行植株；
去除检测模块，设置为根据所述采样图像识别所述采样图像中的全部母本行植株的目标物以进行目标物去除检测。
本申请的实施例提供一种图像采集方法，包括：
获取规则种植的植株区域的采样点的图像信息，所述植株区域包括父本行植株和母本行植株；
根据所述图像信息确定图像采集设备的目标采样姿态，所述目标采样姿态包括以下的至少一种：采集高度和采集角度，其中所述采集角度为所述图像采集设备的图像水平轴与植株种植行的延伸方向的角度；
基于所述目标采样姿态获取所述采样点对应的采样图像。
本申请的实施例还提供一种图像采集装置，包括：
图像获取模块，设置为获取规则种植的植株区域的采样点的图像信息，所述植株区域包括父本行植株和母本行植株；
姿态确定模块，设置为根据所述图像信息确定图像采集设备的目标采样姿态，所述目标采样姿态包括以下的至少一种：采集高度和采集角度，其中所述采集角度为所述图像采集设备的图像水平轴与植株种植行的延伸方向的角度；
图像确定模块，设置为基于所述目标采样姿态获取所述采样点对应的采样图像。
本申请实施例还提供一种电子设备，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上述的目标物检测方法或图像采集方法。
本申请实施例还提供一种计算机可读存储介质，计算机可读存储介质上存储有计算机程序，计算机程序被处理器运行时执行上述的目标物检测方法或图像采集方法。
附图说明
为了说明本申请实施方式或相关技术中的技术方案，下面将对实施方式或相关技术描述中所需要使用的附图作简单地介绍。
图1为本申请实施例提供的一种父本植株和母本植株种植示意图；
图2为本申请实施例提供的一种图像采集方法流程图；
图3为本申请实施例提供的一种图像采集设备处于准备位置的图像信息示意图；
图4为本申请实施例提供的一种图像采集设备处于采样位置的图像信息示意图；
图5为本申请实施例提供的一种目标物检测方法流程图；
图6为本申请实施例提供的一种用于去雄检测的用户界面示意图；
图7为本申请实施例提供的一种去雄检测方法应用场景示意图；
图8为本申请实施例提供的一种图像采集装置的功能模块图；
图9为本申请实施例提供的一种目标物检测装置的功能模块图；
图10为本申请实施例提供的一种电子设备的硬件架构示意图。
具体实施方式
下面将结合附图对本申请的技术方案进行描述，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。
去除检测广泛应用于农业育种领域，以玉米去雄为例进行说明。当制种公司制种时，一般种植区域达1000亩以上，种植区域地块边界达1公里以上，而植株都是成行种植，在去雄时，首先通过机械设备去雄，来去除母本行上的雄穗，保留父本行雄穗用于授粉，但是在机械设备去雄后母本行上往往会存在遗漏未去除的雄穗，或者去雄时未长成但后续生长出的雄穗，此时为了保证母本行的去雄率，通过人力行走、人力检测来抽检这1000亩地的去雄情况，并且需 要间隔一段时间进行多次抽检，不仅花费巨大的时间，也花费巨大的人力物力。
在玉米制种时，采用间行种植，即将父本行和母本行间隔种植，如图1所示。在授粉期间，将母本行的雄穗去除，只留下父本行的雄穗，授粉时，母本行植株上结的果实都是父本的花粉和母本的卵子结合而成，如此实现杂交制种。如果母本行去雄不好，母本花粉授到自身的果穗上，形成自交种子，会极大影响种子纯度。基于此，要保证母本行上的雄穗完全去除，进而需要进行去雄检测，以实现较好的种子纯度。植株的种植规律可如图1中所示，例如：母本行4行，父本行2行，母本行4行，父本行2行间隔种植，并且行距确定；也可以母本行6行，父本行2行，母本行6行，父本行2行间隔种植。这里仅为一种示例，母本行的行数和父本行的行数不做限制，行间距也不做限制。
对于去雄检测的工作人员来说，需要人工行走至采样点，费时费力，玉米等作物的种植区域里情况复杂，可能存在遇到蛇虫，遭遇中暑等危险情况，在通过人工检查是否去雄，效率较低，成本较高的同时，由于受成本限制，不能对植株进行大量取样，导致难以保证种子的纯度。
由于人工去雄检测方法的以上局限性，可利用自动控制设备控制图像采集设备，如飞行器航拍，来获取去雄后的种田图片从而识别去雄率，并通过深度学习的方法进行解读。但是该方案由于存在以下技术难点，故未能被成功运用到生产实践中：
通过机器学习、图像识别难以自动高精度分开父本植株和母本植株；
由于花期玉米枝叶交错，机器学习、图像识别难以自动准确地数清玉米株数从而给出遗漏雄穗占比；
对于离地高度，所含植株数，曝光程度不同的图像难以准确进行识别；
由于田间情况复杂，图片间差异巨大，难以高精度检测大数量种类的玉米 中的遗漏雄穗；
检测流程复杂，人工参与度高，难以真正有效提高去雄检测效率和精度。
本申请实施例提供的一种目标物检测方法、装置和图像采集方法、装置，通过机器控制方式采集高度一致性的采样图像，保证目标物去除检测的效率和准确性，省时省力，又降低了检测成本。
为便于对本实施例进行理解，首先对本申请实施例所公开的一种图像采集方法进行介绍，主要应用于控制设备，如飞行器，适应于规则种植的植株场景中，此处规则种植包括但不限于沿行种植或者小区种植，比如规则种植的玉米、水稻、大豆、油菜等，本申请实施例以玉米去雄为例进行说明。
图2为本申请实施例提供的一种图像采集方法流程图。
参照图2，本申请提供的一种图像采集方法，主要包括以下步骤：
步骤S102，获取规则种植的植株区域的采样点的图像信息，植株区域包括父本行植株和母本行植株。
步骤S104，根据图像信息确定图像采集设备的目标采样姿态，目标采样姿态至少包括以下的一种或多种：采集高度和采集角度，采集角度为图像采集设备的图像水平轴与植株种植行的延伸方向的角度。
步骤S106，基于目标采样姿态获取采样点对应的采样图像。
在实际应用的可选实施例中，根据获取的植株区域的采样点的图像信息确定图像采集设备的采集高度、采集角度达到目标采样姿态，根据目标采样姿态在采样点采集高度一致的采样图像，以根据采样图像中的植株情况进行目标物去除检测，如去雄检测，本申请实施例通过机器控制方式采集高度一致性的采样图像，进而准确识别遗漏雄穗和植株数量，提高了去雄检测的自动化程度，同时保证去雄检测的准确性，省时省力。
本申请实施例以去雄检测为例进行说明，并不局限于此，还适用于其他的目标物去除检测场景。
通过机器控制方式标准化地采集高度一致性的采样图像，主要针对规则种植的植株场景(种植规则一致性较高)，通过本申请的图像采集方法(采集方式高度一致)，进而可以获得高度一致性的采样图像(确保每张采样图像的实际面积相同)，进而采样的植株株数一致性高(一般每张采样图像的植株数相同)，从而可以很好地进行抽样检查。
作为一种可选的实施例，对于种植区域中规则种植的植株，可通过遥控控制设备并且人为确定飞行器拍摄的平面位置以及高度位置，以此保证拍摄图像的一致性，进而排除其他植株行的干扰，提高采样图像的一致性。通过人眼确认父本行和母本行的边界，在采样时排除父本行的干扰的情况下，确定采集高度和采集角度，保证了准确性，避免了由于父本行和母本行的外形极为相似，机器识别难以准确分辨的难点。获得采样图像后，得到一张采样图像中显示的植株株数(可以通过人为计数，也可以通过图像识别，此处不做限制)，就可以得到所有采样图像中的植株总数，以便后续统计。通过图像采集设备采集母本行的雄穗图像检测去雄是否符合要求，提高了检测效率，节省了人力物力。通过遥控飞行器采集雄穗图像进行雄穗抽检，避免人工作业，避免深入农田的风险，节省了人力物力，提高了检测效果和检测效率。
作为另一种可选的实施例，可通过上位机控制飞行器的姿态，即采集高度、采集角度等，进而飞行器搭载图像采集设备(相机)进行图像采集。
在可选的实施方式中，步骤S104，还包括以下步骤：
步骤1.1)，调整图像采集设备的当前高度直至图像信息中包括除父本行植株外的全部母本行植株，确定目标采样姿态中的采集高度。
如图4所示，要求当图像信息中刚刚看不到父本行而保留相邻父本行间的所有母本行时，此时的航拍高度即为目标采样姿态中的采集高度。在玉米去雄的场景中，拍摄的边界以父本植株为边界，在其他连续植株种植行的应用场景中，可以根据需要进行设定。
在可选的实施方式中，目标采样姿态包括相同的采集方向，采集方向为图像采集设备朝向地面的方向，如竖直向下。
作为一种可选实施例，要求航拍镜头方向垂直向下对植株进行拍摄，进而获得采样图像，以便识别雄穗的顶端，提高雄穗的识别率，如果倾斜，会识别到雄穗的部分本体，重叠在一起的雄穗会影响识别准确率。
在可选的实施方式中，步骤S104，还可通过以下步骤进行实现：
步骤1.2)，调整图像采集设备的当前采集角度直至图像采集设备的图像水平轴与所述图像信息中的植株种植行的延伸方向呈预设角度，确定目标采样姿态中的采集角度。获取采样图像时，要求图像信息中的植株种植行的延伸方向与图像水平轴的预设角度不超过±15°，以保证图像识别的准确性和鲁棒性。在一种可选实施例中，调整图像采集设备的当前采集角度直至图像采集设备的图像水平轴平行或垂直于图像信息中的植株种植行的延伸方向，确定目标采样姿态中的采集角度，采集角度为0°或90°。即采集的图像中的植株种植行的延伸方向平行或者垂直于图像边界线，进而图像采集设备的采集角度可选为0°或90°。
在可选的实施方式中，本申请实施例提供的方法还包括：
步骤1.3)，根据图像信息的亮度控制采样图像的曝光度。
根据图像信息中的最亮处不能有白斑的要求，来控制航拍亮度。进而保证采样图像的质量，防止采样图像无法识别的情况。
在可选的实施方式中，植株区域中包括多个采样点，每个采样点对应的所述采样图像互不重叠。
为了较为准确地对植株区域进行去除检测，一般植株区域可包括多个采样点，需要保证多个采样点采集的采样图像互不重叠，避免雄穗数量以及植株数量统计误差，进而确保去雄检测结果的准确性。
作为一种可选的实施例，先控制(飞行器遥控或上位机控制)航拍飞行器飞至要进行采集的目标采样点上方，悬停并调整飞行器的飞行高度，镜头采集角度和采集方向达到上述要求，并且保证目标母本行以及目标母本行两侧的父本行全部位于画面内，如图3所示的航拍飞行器的准备位置。再缓慢降低航拍飞行器的高度并调整飞行器的位置和姿态，直到父本行刚刚移出画面而相邻的父本行中间的所有的母本行完全保留在画面内为止，并且采样图像的画面水平轴平行于植株种植行的延伸方向，至如图4所示的航拍飞行器位置，调节曝光度至画面最亮处没有白斑，进行图像拍摄采集。如果采样图像中仅包含部分母本行，如采样图像中包括3行母本行，或者2行母本行，会影响抽样的准确性。
本申请实施例中的去雄检测采样图像以及深度学习训练集图片都需要依照同样的上述办法进行图像采集，以保证雄穗图像检测准确性。在可选的实施方式中，步骤S102，还包括以下步骤：
步骤2.1)，获取规则种植的植株区域中的采样点以及采样点的位置坐标；
步骤2.2)，根据采样点的位置坐标，规划图像采集设备到采样点的飞行路径；
步骤2.3)，基于飞行路径获得图像信息。
获取采集图像方法的执行主体为飞控(控制搭载图像采集设备的飞行器)或搭载图像采集设备的飞行器，通过控制飞行器达到采样点，控制图像采集设 备采集采样图像来实现。
作为一种可选的实施例，飞行器接收种植区域中的采样点的位置坐标，飞行器根据飞行路径到达该采样点上空；采样点可以是随机的，也可以是预设的。飞行器可以通过操作人员遥控达到多个采样点，或者飞行器根据预先设置或随机设置的种植区域的多个采样点位置规划飞行路径，根据规划的路径自动到达多个采样点进而进行图像采集。
在实际应用的图像采集场景中，还通过降低飞行器飞行高度至预设值，判断(机器判断或者人工判断)图像采集设备拍摄的图像中是否存在父本行植株，此时母本行植株可能存在多个未去除的雄穗，而父本行的植株上雄穗未经过去雄，如果识别到父本行植株，会对识别结果造成影响。同时，若图像中存在父本行，还可以控制飞行器调整采集角度，包括但不限于调整飞行器飞行姿态或者图像采集设备的姿态等等，从而使得图像采集设备的图像水平轴与种植植株行的延伸方向的角度呈0°，直到图像采集设备采集到的都是母本行植株为止，确定该位置为采样位置，确定该采样位置的目标采样姿态。
基于上述方式采集的图像信息中包括采样点的位置坐标，在可选的实施例中，采样图像包括位置信息，如采样点对应的位置信息。基于采样点对应的位置信息，以便定位目标物的准确位置。
在上述实施例的基础上，控制图像采集设备采集该采样点的图像信息，获得采样图像，并进行保存。重复前述步骤操作，直到获得的采样图像的数量符合要求。为了更加准确地获得去雄检测结果，需要采集大量的采样图像。
在一些实施例中，在间隔预设时间后，需要在上述种植区域中重复上述图像采样步骤，该步骤需要重复多次，以保证在植株生长期间，持续去雄检测，防止后长出的雄穗没有检测到。
在确定为采样位置之前，还需要确认拍摄图像中是否包括所有连续种植行的母本行植株(机器判断或者人工判断)，例如，若已知连续种植的母本行的行数为4行，则判断拍摄图像中是否包含四行的母本行，如果仅包含三行，遗漏一行未抽检，这种情况会对抽检结果的正确性造成影响。行数是已知的，可以根据种植规律进行预设。
对于采样图像来说，如图4所示，采样图像为下视图，为镜头竖直向下视角的图像，并且采集不同的图像时图像采集设备镜头的姿态均固定(相同)。可以固定图像采集设备镜头的轴线为竖直方向，在采集图像时，保持飞行器的姿态不变，保证图像采集设备采集图像的姿态相同，也可以通过云台来控制图像采集设备的镜头，随着飞行器姿态的变化而变化，保证图像采集设备镜头在采集图像时姿态相同、采集方向相同。
当实际作业时，如图7所示，操作人员可以遥控飞行器达到植株区域的一个随机或者设定的采样点后，调整采集方向，遥控飞行器降低飞行高度，在高度降低的同时，通过图像采集设备拍摄到的实时图像，确认拍摄图像是否仅包括母本行，并且玉米行的延伸方向平行或垂直于画面水平轴，如果满足条件，并且包含所有连续种植的母本行，则控制图像采集设备采集图像；若拍摄图像还包括父本行或者仅包含少量连续母本行，或者植株种植行的延伸方向不平行或垂直于画面水平轴，在高度降低的同时调整飞行器的位置，直至拍摄图像中仅包含母本行且包含所有的连续母本行后，并且植株种植行的延伸方向平行或垂直于图像水平轴，可以采集该位置的图像。飞行器飞行高度不预先进行设定，而是根据植株的行间距、种植母本行的行数、植株的列间距、航拍图像采集设备视场大小等在拍摄时动态确定的。在实际去雄检测时，图像采集设备可以在采样点获得80张图片，每张图片可包含30颗植株，则一个采样点可以获得2400 颗植株的去雄信息，极大提高了去雄检测的效率，图片的数量和植株的数量不做限制，仅示例。当完成该采样点的采集后，操作人员遥控飞行器达到另一个随机或设定的采样点，重复上述操作，直到完成多个采样点的采集。采样点个数可以为10个，20个，30个，40个，50个，60个，70个，80个，90个，100个等等，不做限制。
作为一种可选的实施例，如果获得了种植区域的位置信息，以及植株的行间距、植株的列间距、种植母本行的行数、种植父本行的行数等等信息，可以规划飞行器采集图片的飞行路径，实现自动采集。此时，先根据预设或者随机的方式确定种植区域内的采样点的位置，规划飞行路径，同时根据采样点的位置信息规划采样路径(包括预设飞行高度)，当飞行器获得该飞行路径和采样路径后，根据飞行路径飞至采样点，并根据每个采样点的采样路径降低至预设高度(计算好的仅能看到母本行并且看到所有连续母本行)，获得采样图像，当完成该采样点的采集后，根据飞行路径飞至下一个采样点，采集图像，直至完成所有采样点的采集后飞回终点。
在一些实施例中，基于前述实施例的图像采集方法获得的一致性较高的采样图像后，可执行目标物检测方法，进而实现对目标物进行去除的目的，如图5所示，包括以下步骤：
步骤S202，获取规则种植的植株区域的采样点的采样图像，植株区域包括父本行植株和母本行植株，采样图像从植株区域的正上方采集，采样图像包括除父本行植株外的全部母本行植株；
步骤S204，根据采样图像识别采样图像中的全部母本行植株的目标物以进行目标物去除检测。
通过目标物检测方法，基于高度一致的采样图像，动态调节图像采集设备 的采样姿态以采集符合要求的采样图像，标注历史采样图像并训练深度学习模型以准确识别目标物，利用训练成功的深度学习模型检测采样图像以获取目标物的数量及比例，从而确定该植株区域的目标物去除程度，并由此保证种子纯度。本申请通过机器取代人工进行自动地样本采集，目标物识别和去除程度统计，大幅提高目标物去除检测效率，大幅缩短目标物去除检测时间，大幅降低目标物检测成本，并彻底避免由于目标物去除检测人员必需深入田间工作时所带来的一切风险。
在实际应用的可选实施例中，通过从植株区域的正上方采集的包括有除父本行植株外的全部母本行植株的采样图像，来进行目标物识别，进而实现目标物去除情况的检测。
在可选的实施例中，采样图像的水平轴与植株种植行的延伸方向呈预设角度。以便提高目标物检测准确度，防止漏检。
在可选的实施例中，植株区域中包括多个采样点，多个采样点对应的采样图像互不重叠。进一步提高目标物识别检测效率和准确度。
在可选的实施例中，采样图像包括位置信息，以便后续定位目标物。
在可选的实施例中，目标物包括雄穗和花苞中的任一种或两种。在玉米去雄检测中，既可以检测雄穗、也可以检测花苞，花苞的检测难度远远大于雄穗，故可以作为一种参考检测方案。
满足前述实施例的采样图像具有更高的一致性，进而在图像识别过程中，能够更加准确地对目标物去除情况进行检测，获得更加准确的检测结果。
一般的，飞行器图像采集设备在采集到采样图像后，存储于存储器中以便在采样结束后，导入预设识别模型，进行识别处理，降低飞行器成本；当然，也可以在飞行器上设置计算处理设备，飞行时实时计算处理采样图像，快速获 得识别结果。
本申请实施例提供的方法通过规定目标物采集的高度、角度和方向来采集采样图像，保证了获得图像的一致性，从而提高了识别精确性；同时通过控制预设飞行高度来排除父本行的干扰，可以保证识别所有连续种植的母本行，避免母本行的遗漏，进一步提高识别检测精确性。
在可选的实施例中，步骤S204还可用以下步骤实现：
步骤3.1)，根据历史采样图像训练深度学习模型，以获得目标物模型。
可通过标注历史采样图像形成已标注的采样图像的训练集，以训练深度学习模型，进而获得目标物模型。
步骤3.2)，根据目标物模型识别采样图像中的全部母本行植株的目标物。
进而，可根据目标物模型识别未标注的采样图像中的全部母本行植株的目标物。
在可选的实施例中，深度学习模型包括任意开源或自行开发的基于深度学习的目标物检测的神经网络。基于同一标准的采样图像以训练深度学习模型、以及识别检测目标物，进一步提高目标物检测的准确性。
在可选的实施例中，步骤S204，还包括：
步骤4.1)，确定采样图像中的全部母本行植株的植株数量；
作为一种可选的实施例，获得的多个采样图像的植株数量分别为c1，c2，c3….cn，该植株数量是已知的，可以是自动识别的，或者是手动人工读取的，求得采样点所有的植株总数c＝c1+c2…+cn，如果采样图像的大小一致，则获得的每个采样图像的植株数量为c0，以及采样图像的数量N，则c＝c0*N。
作为一种可选的实施例，在种植区域中选择一采样点，进行图像采集，采集完成后，到达下一个采样点，再次进行图像采集，如此重复，抽检整个种植 区域内的目标物去除图像信息。可以随机选择采样点，可以根据种植区域预先确定采样点，不做限制。采样点的数量不做限制，为了提高检测准确性，可以适当提高采样点的数量。采样点的数量还可以根据图像采集设备的分辨率、飞行高度、种植行间距、种植列间距等来确定，以提高抽检准确率。
采样图像可以存储于飞行器的存储器中，也可以实时回传，不做限制。采样图像仅包括母本行植株，排除父本行植株，提高识别准确度，避免了父本行的干扰。同时，采样图像包括所有的连续种植行的母本行植株，避免遗漏，防止漏检。并且，采样图像包括的植株数量大于或等于预设数值，避免植株数量过小而检测效率低，同时植株数量也小于第二预设数值，避免分辨率过低而检测不准确，保证抽样统计的准确性。图像可以由测绘设备、照相机等图像采集设备提供，图像包括测绘图像信息、图片信息中的一种或多种，不仅限于此。
步骤4.2)，根据识别的目标物统计目标物数量。
可根据深度学习模型识别采样图像中的雄穗纹理特征，得到采样图像中的雄穗数量；利用大量的历史数据(训练图像)对预设的识别模型进行训练，获得雄穗的深度学习模型，该深度学习模型通过训练图像对采样图像中的雄穗进行识别，这里的训练图像的采集方式与前述实施例中采样图像的采集方式相同，在此不再赘述，训练图像和采样图像保持一致，提高检测精度。
将采样图像输入深度学习模型，深度学习模型对采样图像进行检测，得到雄穗信息。从获得的采样图像中提取待测雄穗图形特征，利用深度学习模型对待测雄穗图形特性进行处理，获得雄穗信息。可以一次性获得多个采样点的采样图像后，在专门的计算处理设备上进行图像处理；也可以每次获得采样图像时，实时处理，这取决于飞行器是否配备计算处理器，以及计算处理器的计算能力。
步骤4.3)，根据植株数量和目标物数量确定目标物去除程度。
可根据植株总量和雄穗数量确定去雄率；进而判断去雄率是否达到去雄率阈值；若去雄率达到去雄率阈值，则去雄合格；若去雄率未达到去雄率阈值，则去雄不合格。
作为一种可选的实施例，通过手动人工读取的方式，求得采样点所有的植株总数c，此时识别到的雄穗总数为b＝b1+b2…+bn，则去雄率d计算方式为d＝b/c，可以根据d的数值大小来确定是否已经完成去雄。比如，预设阈值为确保种子达到纯度要求的雄穗比例，当d小于或等于预设阈值时，则表示符合去雄的要求，无需再次去雄，当d大于或等于预设阈值时，还需要再次去雄。在间隔预设时间后，在同一种植区域重复前述采样图像采集步骤，此时获得同一个种植区域的采样图像，该采样图像反映了植株生长了一段时间后的雄穗情况，在原来的采样点或者同一种植区域的其他采样点重新采集，避免后长出的雄穗未检测到，并在此基础上，重复多次，保证在一段时间的生长期内，所有生长的雄穗都能检测到，都符合去雄要求。
本申请所采集的采样图像具有分辨率高，一致性好的优点。高分辨率可以保持雄穗的纹理特征，高一致性使训练集和检测集的采样图像特征更加接近，这样运用成熟的深度学习目标检测技术，在对大量的历史采样图像的训练的基础上，可以很容易提取采样图像特征，识别雄穗，由计算机自动获取每张采样图像里的雄穗个数。同时由于每张采样图像里只有母本行，从而完全避免了父本对雄穗计数的干扰。
根据采集到的母本行植株图像(采样图像)中识别到的雄穗数量，对雄穗的高精度识别计数，计算去雄率。本申请实施例中的去雄纯度估计方法，在以上的植株计数方法和高精度玉米雄穗识别计数方法的基础上，通过如下公式即 可得到所检测植株区域的玉米去雄纯度估计值：
去除率＝检测到的所有采样图像里的总目标物数/(采样图像总数*平均每张采样图像里的植株株数)
该去除率可以按地块按时间(天)分别计算，从而发现变化趋势，上述母本行植株图像为种植区域中随机区域(采样点)的图像或者预设区域(采样点)的图像，数量不做限制。一般来说，数量越多，获得的去除率越准确。
在可选的实施例中，步骤4.1)，还可包括以下步骤：
步骤4.1.1)，对采样图像中的全部母本行植株的植株数量进行统计；
或者，
步骤4.1.2)，对采样图像中的全部母本行植株进行识别，得到植株数量；
或者，
步骤4.1.3)，获取预设数量的采样图像中的全部母本行植株的植株数量，计算预设数量的采样图像中的全部母本行植株的植株平均数量，根据植株平均数量与采样图像的总数量确定采样图像中的全部母本行植株的植株数量。
根据采样图像能够获得在种植区域采集到的母本行植株图像中的植株总数，仅包括母本行，并且包括所有连续种植行的母本行。
由于根据上述实施例的采样图像具有高度一致性，而同一制种田块的植株的种植距离基本固定，每张采样图像里的植株数基本一致，从而可以从多张采样图像提前手工计数获取每张采样图像的平均植株数。该图像-植株数对映关系一般根据田块和植株种类而变化。对于一个田块，植株种类一般只需要计数一次即可，从而避免了人工识图计数的高成本或机器识图计数带来的准确性问题；还可以通过机器自动识别采样图像来获得植株数；或者手工对每张采样图像计数，统计获得植株数。
作为一种可选的实施例，对每一地块或同一种植方式和密度的地块集合，随机抽取多张采样图像，人工数出多张采样图像所含的植株数并计算平均每张采样图像所含植株数C，此时通过目标物检测方法检测出每张采样图像i所含的目标物数量，雄穗数Ti和花苞数Bi(i＝1，2…N，N为该地块所获得的采样图像数量)。
则该地块的遗漏雄穗比例为：Rt＝Sum(Ti)/(N*C)；
遗漏花苞比例为：Rb＝Sum(Bi)/(N*C)；
目标物去除程度为：D＝1-(Rt+Rb)。
目标物检测不限于一种，可以在同一地块上同时检测多种不同的目标物，获得彻底去除程度结果。
本申请实施例提供的目标物检测方法利用预设深度学习模型对采样图像进行处理，获得采样点的目标物信息，简化了目标物的统计过程，从而提高目标物的统计效率，避免了人工统计种植区域的目标物数据所导致的统计效率低下，以及统计过程繁复的问题。
作为另一种可能的实施例，如图6所示，操作人员可通过用户界面实现去雄检测，去雄检测系统用户界面包括用户登录及管理系统、图片上传及管理系统、人工智能(Artificial Intelligence，AI)雄穗识别系统、识别结果呈现及报表系统、用户反馈系统、面积及计费系统，AI雄穗识别模型、去雄检测数据库和去雄检测图片文件系统，可实现对采样图像中雄穗和植株进行识别，进而计算去除率的操作。
在一些实施例中，计算去除率后，还可通过抽检的方式自动检测植株种植区域中是否存在目标物，判断是否符合去除要求。
本申请实施例基于深度学习和航拍图片实现目标物去除检测，着眼于解决 以下几个问题：1.父本和母本识别问题；2.识别精度问题；3.目标物比例问题；4.操作效率问题。通过本申请的目标物检测方法，控制飞行器航拍飞行高度，拍摄角度，拍摄方向，感光程度等采集姿态，从而获取标准化的高度一致性的航拍采样图片，实现植株株数计数，并且降低采样图像的识别难度，提高检测精度，达到准确高效的目标物去除检测效果。
如图8所示，实施例还提供一种图像采集装置800，包括：
图像获取模块801，设置为获取规则种植的植株区域的采样点的图像信息，植株区域包括父本行植株和母本行植株；
姿态确定模块802，设置为根据图像信息确定图像采集设备的目标采样姿态，目标采样姿态包括以下的至少一种：采集高度和采集角度，采集角度为图像采集设备的图像水平轴与植株种植行的延伸方向的角度；
图像确定模块803，设置为基于目标采样姿态获取采样点对应的采样图像，其中，针对采样图像中的植株进行目标物去除检测。
在可选的实施方式中，所述采样图像包括位置信息。
在可选的实施方式中，姿态确定模块802是设置为调整图像采集设备的当前高度直至图像信息中包括除父本行植株外的全部母本行植株，确定目标采样姿态中的采集高度。
在可选的实施方式中，目标采集姿态包括采集方向，采集方向为图像采集设备朝向地面的方向，其中，采集方向为竖直向下。
在可选的实施方式中，姿态确定模块802是设置为调整图像采集设备的当前采集角度直至图像采集设备的图像水平轴与所述图像信息中的植株种植行的延伸方向呈预设角度，确定目标采样姿态中的采集角度。
在可选的实施方式中，图像获取模块801是设置为获取规则种植的植株区 域中的采样点以及采样点的位置坐标；根据采样点的位置坐标，规划图像采集设备到采样点的飞行路径；基于飞行路径获得图像信息。
在可选的实施方式中，图像采集装置还包括根据图像信息的亮度控制采样图像的曝光度。
如图9所示，实施例还提供一种目标物检测装置900，包括：
采样图像获取模块901，设置为获取规则种植的植株区域的采样点的采样图像，所述植株区域包括父本行植株和母本行植株，所述采样图像从所述植株的正上方采集，所述采样图像包括除父本行植株外的全部母本行植株；
去除检测模块902，设置为根据所述采样图像识别所述采样图像中的全部母本行植株的目标物以进行目标物去除检测。
在可选的实施方式中，所述采样图像的水平轴与植株种植行的延伸方向呈预设角度。
在可选的实施方式中，植株区域中包括多个采样点，多个采样点对应的采样图像互不重叠。
在可选的实施方式中，所述采样图像包括位置信息。
在可选的实施方式中，所述目标物包括雄穗和花苞中的至少之一。
在可选的实施方式中，去除检测模块902是设置为根据所述采样图像训练深度学习模型，以获得目标物模型；根据所述目标物模型识别所述采样图像中的全部母本行植株的目标物。
在可选的实施方式中，所述深度学习模型包括任意开源或自行开发的基于深度学习的目标物检测的神经网络。
在可选的实施方式中，去除检测模块902是设置为通过如下方式根据所述采样图像训练深度学习模型，以获得目标物模型：标注所述采样图像形成已标 注的采样图像的训练集，根据所述训练集训练深度学习模型，以获得目标物模型；去除检测模块902是设置为通过如下方式根据所述目标物模型识别所述采样图像中的全部母本行植株的目标物：根据所述目标物模型识别未标注的采样图像中的全部母本行植株的目标物。
在可选的实施方式中，去除检测模块902还设置为确定所述采样图像中的全部母本行植株的植株数量；根据识别的目标物统计目标物数量；根据所述植株数量和所述目标物数量确定目标物去除程度。
在可选的实施方式中，去除检测模块902是设置为通过如下方式确定所述采样图像中的全部母本行植株的植株数量：对采样图像中的全部母本行植株的植株数量进行统计；或者，对采样图像中的全部母本行植株进行识别，得到植株数量；或者，获取预设数量的采样图像中的全部母本行植株的植株数量，计算预设数量的采样图像中的全部母本行植株的植株平均数量，根据植株平均数量与采样图像的总数量确定采样图像中的全部母本行植株的植株数量。
图10为本申请实施例提供的电子设备1000的硬件架构示意图。参见图10所示，该电子设备包括：机器可读存储介质1001和处理器1002，还可以包括非易失性存储介质1003、通信接口1004和总线1005；机器可读存储介质1001、处理器1002、非易失性存储介质1003和通信接口1004通过总线1005完成相互间的通信。处理器1002通过读取并执行机器可读存储介质1001中的目标物检测的机器可执行指令，可执行上文实施例描述目标物检测方法或图像采集方法。
本文中提到的机器可读存储介质可以是任何电子、磁性、光学或其它物理存储装置，可以包含或存储信息，如可执行指令、数据，等等。例如，机器可读存储介质可以是：随机存取存储器(Radom Access Memory，RAM)、易失存储器、非易失性存储器、闪存、存储驱动器(如硬盘驱动器)、任何类型的存 储盘(如光盘、数字通用光盘(Digital Versatile Disc，DVD)等)，或者类似的存储介质，或者它们的组合。
非易失性介质可以是非易失性存储器、闪存、存储驱动器(如硬盘驱动器)、任何类型的存储盘(如光盘、DVD等)，或者类似的非易失性存储介质，或者它们的组合。
本实施例中的多个功能模块的操作方法可参照上述方法实施例中相应步骤的描述，在此不再重复赘述。
本申请实施例所提供计算机可读存储介质，所述可读存储介质中存储有计算机程序，所述计算机程序代码被执行时可实现上述任一实施例所述的目标物检测方法或图像采集方法，实现可参见方法实施例，在此不再赘述。
所属领域的技术人员可以了解到，为描述的方便和简洁，上述描述的装置的工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。

Claims (21)

  1. 一种目标物检测方法，包括：
    获取规则种植的植株区域的采样点的采样图像，所述植株区域包括父本行植株和母本行植株，所述采样图像从所述植株区域的正上方采集，所述采样图像包括除父本行植株外的全部母本行植株；
    根据所述采样图像识别所述采样图像中的全部母本行植株的目标物以进行目标物去除检测。
  2. 根据权利要求1所述方法，其中，所述采样图像的水平轴与植株种植行的延伸方向呈预设角度。
  3. 根据权利要求1所述方法，其中，所述植株区域中包括多个采样点，多个采样点对应的采样图像互不重叠。
  4. 根据权利要求1所述方法，其中，所述采样图像包括位置信息。
  5. 根据权利要求1所述方法，其中，所述目标物包括雄穗和花苞中的至少之一。
  6. 根据权利要求1所述方法，其中，根据所述采样图像识别所述采样图像中的全部母本行植株的目标物以进行目标物去除检测，包括：
    根据历史采样图像训练深度学习模型，以获得目标物模型；
    根据所述目标物模型识别所述采样图像中的全部母本行植株的目标物。
  7. 根据权利要求6所述方法，其中，所述深度学习模型包括任意开源或自行开发的基于深度学习的目标物检测的神经网络。
  8. 根据权利要求6所述方法，其中，根据历史采样图像训练深度学习模型，以获得目标物模型，包括：
    标注历史采样图像形成已标注的采样图像的训练集，根据所述训练集训练深度学习模型，以获得目标物模型；
    根据所述目标物模型识别所述采样图像中的全部母本行植株的目标物，包括：根据所述目标物模型识别未标注的采样图像中的全部母本行植株的目标物。
  9. 根据权利要求6所述方法，其中，根据所述采样图像识别所述采样图像中的全部母本行植株的目标物以进行目标物去除检测，还包括：
    确定所述采样图像中的全部母本行植株的植株数量；
    根据识别的所述目标物统计目标物数量；
    根据所述植株数量和所述目标物数量确定目标物去除程度。
  10. 根据权利要求9所述方法，其中，确定所述采样图像中的全部母本行植株的植株数量，包括：
    对所述采样图像中的全部母本行植株的植株数量进行统计；
    或者，
    对所述采样图像中的全部母本行植株进行识别，得到植株数量；
    或者，
    获取预设数量的采样图像中的全部母本行植株的植株数量，计算所述预设数量的采样图像中的全部母本行植株的植株平均数量，根据所述植株平均数量与所述采样图像的总数量确定所述采样图像中的全部母本行植株的植株数量。
  11. 一种目标物检测装置，包括：
    采样图像获取模块，设置为获取规则种植的植株区域的采样点的采样图像，所述植株区域包括父本行植株和母本行植株，所述采样图像从所述植株区域的正上方采集，所述采样图像包括除父本行植株外的全部母本行植株；
    去除检测模块，设置为根据所述采样图像识别所述采样图像中的全部母本行植株的目标物以进行目标物去除检测。
  12. 一种图像采集方法，包括：
    获取规则种植的植株区域的采样点的图像信息，所述植株区域包括父本行植株和母本行植株；
    根据所述图像信息确定图像采集设备的目标采样姿态，所述目标采样姿态包括以下的至少一种：采集高度和采集角度，所述采集角度为所述图像采集设备的图像水平轴与植株种植行的延伸方向的角度；
    基于所述目标采样姿态获取所述采样点对应的采样图像。
  13. 根据权利要求12所述方法，其中，所述采样图像包括位置信息。
  14. 根据权利要求12所述方法，其中，根据所述图像信息确定图像采集设备的目标采样姿态，包括：
    调整所述图像采集设备的当前高度直至所述图像信息中包括除父本行植株 外的全部母本行植株，确定所述目标采样姿态中的采集高度。
  15. 根据权利要求12所述方法，其中，根据所述图像信息确定图像采集设备的目标采样姿态，包括：
    调整所述图像采集设备的当前采集角度直至所述图像采集设备的图像水平轴与所述图像信息中的植株种植行的延伸方向呈预设角度，确定所述目标采样姿态中的采集角度。
  16. 根据权利要求12所述方法，其中，所述目标采集姿态还包括采集方向，所述采集方向为图像采集设备朝向地面的方向，其中，所述采集方向为竖直向下。
  17. 根据权利要求12所述方法，其中，获取规则种植的植株区域的采样点的图像信息，包括：
    获取规则种植的植株区域中的采样点以及所述采样点的位置坐标；
    根据所述采样点的位置坐标，规划所述图像采集设备到所述采样点的飞行路径；
    基于所述飞行路径获得所述图像信息。
  18. 根据权利要求12所述方法，还包括：
    根据所述图像信息的亮度控制所述采样图像的曝光度。
  19. 一种图像采集装置，包括：
    图像获取模块，设置为获取规则种植的植株区域的采样点的图像信息，所述植株区域包括父本行植株和母本行植株；
    姿态确定模块，设置为根据所述图像信息确定图像采集设备的目标采样姿态，所述目标采样姿态包括以下的至少一种：采集高度和采集角度，其中所述采集角度为所述图像采集设备的图像水平轴与植株种植行的延伸方向的角度；
    图像确定模块，设置为基于所述目标采样姿态获取所述采样点对应的采样图像。
  20. 一种电子设备，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权利要求1-10中任一所述的目标物检测方法或权利要求12-18中任一所述的图像采集方法。
  21. 一种计算机可读存储介质，存储有计算机程序，计算机程序被处理器运行时执行权利要求1-10中任一所述的目标物检测方法或权利要求12-18中任一所述的图像采集方法。
PCT/CN2020/125251 2020-04-16 2020-10-30 目标物检测方法、装置和图像采集方法、装置 WO2021208407A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
BR112022020889A BR112022020889A2 (pt) 2020-04-16 2020-10-30 Métodos de detecção de objeto alvo e de aquisição de imagem, e, dispositivo eletrônico

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010302566.5 2020-04-16
CN202010302566.5A CN112507768A (zh) 2020-04-16 2020-04-16 目标物检测方法、装置和图像采集方法、装置

Publications (1)

Publication Number Publication Date
WO2021208407A1 true WO2021208407A1 (zh) 2021-10-21

Family

ID=74953230

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/125251 WO2021208407A1 (zh) 2020-04-16 2020-10-30 目标物检测方法、装置和图像采集方法、装置

Country Status (3)

Country Link
CN (1) CN112507768A (zh)
BR (1) BR112022020889A2 (zh)
WO (1) WO2021208407A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114631478A (zh) * 2022-03-25 2022-06-17 安徽科技学院 一种玉米育种田间自交果穗计数装置与控制方法

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113435649B (zh) * 2021-06-29 2022-09-16 布瑞克农业大数据科技集团有限公司 一种全域农业数据整理方法、系统、装置及介质
CN113819921A (zh) * 2021-10-27 2021-12-21 苏州极目机器人科技有限公司 一种执行终端的导航方法及电子设备
CN114489113B (zh) * 2021-12-15 2024-02-23 北京市农林科学院智能装备技术研究中心 一种去雄无人机控制方法及系统
CN114402995B (zh) * 2022-01-19 2023-03-24 北京市农林科学院智能装备技术研究中心 空地协同的玉米去雄方法、系统及空中无人去雄机
CN114766347A (zh) * 2022-04-26 2022-07-22 新疆荣耀九天科技有限公司 一种玉米抽雄作业的目标三维位置获取方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107480706A (zh) * 2017-07-24 2017-12-15 中国农业大学 一种制种玉米田遥感识别方法及装置
CN109324051A (zh) * 2018-11-08 2019-02-12 北方民族大学 一种植物水分检测方法及系统
CN209086157U (zh) * 2018-11-08 2019-07-09 北方民族大学 一种植物水分检测系统
CN110414491A (zh) * 2019-08-29 2019-11-05 新疆农业科学院经济作物研究所 一种棉花单株形态照片的拍摄方法及其辅助装置
US20190339242A1 (en) * 2018-05-06 2019-11-07 Beijing Normal University Method and system for identifying plant species based on hyperspectral data
CN110852341A (zh) * 2019-09-23 2020-02-28 平安科技（深圳）有限公司 基于深度学习的白术检测方法、及其相关设备

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019127395A1 (zh) * 2017-12-29 2019-07-04 深圳市大疆创新科技有限公司 一种无人机拍照方法、图像处理方法和装置
CN110992325A (zh) * 2019-11-27 2020-04-10 同济大学 基于深度学习的目标清点方法、装置和设备

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107480706A (zh) * 2017-07-24 2017-12-15 中国农业大学 一种制种玉米田遥感识别方法及装置
US20190339242A1 (en) * 2018-05-06 2019-11-07 Beijing Normal University Method and system for identifying plant species based on hyperspectral data
CN109324051A (zh) * 2018-11-08 2019-02-12 北方民族大学 一种植物水分检测方法及系统
CN209086157U (zh) * 2018-11-08 2019-07-09 北方民族大学 一种植物水分检测系统
CN110414491A (zh) * 2019-08-29 2019-11-05 新疆农业科学院经济作物研究所 一种棉花单株形态照片的拍摄方法及其辅助装置
CN110852341A (zh) * 2019-09-23 2020-02-28 平安科技（深圳）有限公司 基于深度学习的白术检测方法、及其相关设备

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114631478A (zh) * 2022-03-25 2022-06-17 安徽科技学院 一种玉米育种田间自交果穗计数装置与控制方法

Also Published As

Publication number Publication date
BR112022020889A2 (pt) 2022-11-29
CN112507768A (zh) 2021-03-16

Similar Documents

Publication Publication Date Title
WO2021208407A1 (zh) 目标物检测方法、装置和图像采集方法、装置
CN106971167B (zh) 基于无人机平台的作物生长分析方法及其分析系统
AU2012286783B2 (en) Plant growth kinetics captured by motion tracking
GB2598012A (en) System and method for crop monitoring
US10127450B2 (en) Plot gap identification
CN104732564B (zh) 一种玉米叶面积无损动态监测装置与方法
KR20140077513A (ko) 영상을 통한 온실 작물 정보 관리 시스템 및 그 방법
CN115272187A (zh) 一种基于车载动态视场帧间关联性的田间稻麦倒伏全局评估方法
CN112580671A (zh) 一种基于深度学习的稻穗多发育期自动检测方法及系统
US20180121725A1 (en) Plot gap identification
CN109598215A (zh) 一种基于无人机定位拍摄的果园建模分析系统和方法
JP2023010952A (ja) 農作物の成長監視システム、及び成長監視方法
CN110689022B (zh) 基于叶片匹配的各株作物图像提取方法
Gao et al. Maize seedling information extraction from UAV images based on semi-automatic sample generation and Mask R-CNN model
WO2022104867A1 (zh) 目标物的特征检测方法和装置
CN113920190A (zh) 银杏花穗定向方法及系统
CN117197595A (zh) 基于边缘计算的果树生育期识别方法、装置及管理平台
Saeed et al. Cotton plant part 3D segmentation and architectural trait extraction using point voxel convolutional neural networks
CN117115811A (zh) 一种不依赖无人机的薯类作物垄线高精度确定方法
CN113554691B (zh) 一种植株高度测量方法
CN115527192A (zh) 一种油菜苗期杂株定位方法及杂株去除方法
CN109934186A (zh) 一种农作物直播成苗率的智慧高效监控方法
Bonaria Grapevine yield estimation using image analysis for the variety Arinto
Chimwai et al. Fresh Weight Estimation of Lettuce Using 3D Stereoscopic Technique
Xu et al. TIPS: A three-dimensional phenotypic measurement system for individual maize tassel based on TreeQSM

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: 20930862

Country of ref document: EP

Kind code of ref document: A1

REG Reference to national code

Ref country code: BR

Ref legal event code: B01A

Ref document number: 112022020889

Country of ref document: BR

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 112022020889

Country of ref document: BR

Kind code of ref document: A2

Effective date: 20221014

122 Ep: pct application non-entry in european phase

Ref document number: 20930862

Country of ref document: EP

Kind code of ref document: A1