WO2022104866A1 - 目标物父本处理方法和目标物母本检测方法 - Google Patents

目标物父本处理方法和目标物母本检测方法 Download PDF

Info

Publication number
WO2022104866A1
WO2022104866A1 PCT/CN2020/131745 CN2020131745W WO2022104866A1 WO 2022104866 A1 WO2022104866 A1 WO 2022104866A1 CN 2020131745 W CN2020131745 W CN 2020131745W WO 2022104866 A1 WO2022104866 A1 WO 2022104866A1
Authority
WO
WIPO (PCT)
Prior art keywords
parent
row
target
picture
target object
Prior art date
Application number
PCT/CN2020/131745
Other languages
English (en)
French (fr)
Inventor
陈洪生
王成达
刘叶青
张剑
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
Publication of WO2022104866A1 publication Critical patent/WO2022104866A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/188—Vegetation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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

Definitions

  • the present application relates to the technical field of hybrid breeding, for example, to a method for processing the male parent of a target and a method for detecting the female parent of the target.
  • the row ratio mode is often used, that is, the male parent row and the female parent row are planted at intervals. Because the recognized images often include both male and female parent plants, it is impossible to accurately identify whether the male parent plant or the female parent plant currently has tassel characteristics based on the deep learning algorithm alone, and thus cannot accurately detect the female parent tassel. trait or paternal tassel trait. Therefore, the above identification methods cannot accurately detect the removal of female parent tassel.
  • the present application provides a method for processing a male parent of a target and a method for detecting a female parent of a target. By hiding the male parent row, the purpose of detecting the female parent tassel is achieved and the detection accuracy is improved.
  • an embodiment of the present application provides a method for processing a parent of a target object, including:
  • Hide the parent row and determine the plant picture after hiding the parent row.
  • the embodiments of the present application further provide a method for detecting the parent of a target, including the method for processing the male of the target as described in any one of the above, and the method further includes:
  • the target of the female parent row is determined based on the picture of the plant after hiding the male parent row.
  • FIG. 1 is a flowchart of a method for processing a parent of a target object provided by an embodiment of the present application
  • Fig. 2 is a kind of plant schematic diagram of row ratio planting provided by the embodiment of the application;
  • FIG. 3 is a schematic diagram of an image collected by an unmanned aerial vehicle according to an embodiment of the present application.
  • FIG. 4 is a schematic diagram of a branching operation based on a plant image provided by an embodiment of the present application.
  • FIG. 5 is a schematic diagram of another branching operation based on plant images provided by an embodiment of the present application.
  • FIG. 6 is a schematic diagram of a rotating image provided by an embodiment of the present application.
  • FIG. 7 is a schematic diagram of a plant picture provided in the embodiment of the present application.
  • FIG. 8 is a schematic diagram of a detection interface provided by an embodiment of the present application.
  • Maize hybrid seed production generally adopts the "row ratio mode", in which the parent and the female parent are cross-planted in a certain proportion. That is to say, the female parent row is planted several rows in a row, the male parent row is planted several rows in a row, and then the female parent row continues to plant several rows in a row, according to the method of spacing between the parent and the female parent.
  • Hybrid breeding adopts tasseling treatment, that is, during pollination, the tassel of the female parent row is removed, leaving only the male parent row of the tassel.
  • the fruit on the plant of the female parent row is the pollen of the male parent. It is combined with the eggs of the female parent to achieve hybrid seed production, thereby increasing the nutrients of the corn kernels on the female parent row and increasing the yield.
  • the pollen of the female parent will be pollinated on its own ear to form self-inbred seeds, which will greatly affect the purity of the seeds.
  • Seed production companies require more than 99.7% of seed purity when cultivating seeds. In order to obtain high-purity seeds, seed companies need to completely remove the tassel from the female maize parent. At present, mechanical or artificial emasculation is used, and at the same time of emasculation, the effect of emasculation is detected in time to ensure the emasculation rate, and Re-emasculation when the purity does not meet the requirements. In one embodiment, by analyzing the pictures of the land after emasculation taken by the drone to perform emasculation purity detection by means of artificial intelligence image recognition, manpower and time can be greatly saved and the emasculation detection efficiency can be improved.
  • each picture needs to be manually screened to confirm the removal of the female parent tassel, the workload and the download of the picture are large, and the efficiency is extremely low ; In another embodiment, manually go to the field to check whether the emasculation is complete, but this method is less efficient, the operating conditions are harsh, and the labor time and other costs are high.
  • the plant image is obtained by taking pictures of the plot after emasculation by drone.
  • the plant image generally includes the male parent row and the female parent row.
  • the male parent row retains the male parent row.
  • the female parent row has been emasculated, and the tassel of the female parent row may not be completely emasculated or missed.
  • the above-mentioned image recognition methods cannot quickly and effectively detect the emasculation of the female parent row.
  • the embodiments of the present application provide a method for processing male parent of a target object and a method for detecting female parent of a target object.
  • a method for processing male parent of a target object By hiding the male parent row, the purpose of detecting the female parent row tassel after emasculation is achieved, and the detection accuracy is improved.
  • FIG. 1 is a flowchart of a method for processing a parent of a target object provided by an embodiment of the present application.
  • the target object parent processing method includes the following steps:
  • Step S102 obtaining plant pictures collected in the target plot, wherein the plants in the plant pictures are planted along the rows to form a male parent row and a female parent row, and the plant pictures include multiple, and the multiple plant pictures cover the entire plant.
  • the target plot is described in order to improve the detection accuracy.
  • the plant picture includes plants in the male parent row and the female parent row arranged in a row ratio, the black filled plants are the male parent row plants, and the hollow ones are the female parent row plants.
  • the tasseling operation in the breeding field needs to remove the tassel on all the female parent row plants, and retain the tassel on the male parent row plant, so it is necessary to detect the tassel removal rate of the female parent row.
  • the collected plant pictures can be dozens to hundreds or thousands of pictures covering all planting plots, or collecting plant pictures covering all sampling plots in several sampling plots of all planting plots. That is, in order to improve the detection efficiency, the target plot is a sampling plot. By sampling and selecting several target plots in the whole plot, the number of plant pictures covering all the planting plots is reduced, and the detection efficiency is improved.
  • each plant picture is collected by the same method, which facilitates subsequent recognition by the same model and improves the detection accuracy.
  • the plant pictures are taken and collected at a fixed distance directly above the plants, and the consistency of the posture and method of collecting the images can ensure the consistency of subsequent recognition and reduce the difficulty of model training.
  • the collected plant pictures do not need coordinate information, and the subsequent steps can also be implemented.
  • the collected plant picture has coordinate information, and the identified female parent target (the female parent tassel) can be individually processed according to the coordinate information.
  • Step S104 Determine the parent row in the plant picture according to the branch detection method.
  • the box in the plant row in FIG. 2 is a picture of the plant collected by the camera, including a male parent row and a female parent row, and the male parent row and the female parent row are parallel to the picture boundary line.
  • the box in the plant row is a picture of the plant collected by the camera, the male parent row and the female parent row are not parallel to the picture boundary lines, and the lengths of the male parent row and the female parent row are inconsistent,
  • the male parent row is shorter, but there are more tassel, where the star represents the presence of tassel, the hollow star represents the presence of the female parent row tassel, and the solid star represents the presence of the male parent row tassel.
  • the branch detection method can adapt to different plant pictures in order to effectively identify the parent row.
  • step S104 includes:
  • Step 1.1 divide the plant picture into a plurality of unit lines through the branching operation, determine the number of objects in each unit line of the current picture, and continue to rotate the plant picture at a preset angle after rotating the rotated plant through the branching operation.
  • the picture is divided into a plurality of unit lines, and the number of objects in each unit line of the current picture is determined until the rotation angle of the plant picture reaches 180°.
  • a branch line is determined based on a preset direction and a preset row spacing, and a branch operation is performed on the plant picture based on the branch line to obtain a plurality of unit rows.
  • the plant image is divided into rows with a fixed preset row spacing. As shown in Figure 5, the number of target tassel in rows 1-10 of this unit are: (0, 0, 1, 0, 0, 0, 0 , 0, 1, 1).
  • the preset line spacing is an appropriate spacing, which needs to be determined according to the spacing of the actual parent line and/or the parent line, so as to improve the detection accuracy. If the preset row spacing is too large, there may be 2 rows or different rows of plants in the unit row of the branch, resulting in false detection.
  • the number of target tassel in unit row 1-4 is: (1, 0 , 0, 2), the unit row is too large, resulting in the presence of plants belonging to two rows in unit row 4, which is easy to detect inaccurate; if the preset row spacing is too small, the row cannot be accurately detected When the line is planted crooked, it is easy to not detect the line.
  • Fig. 6 is a picture of the plant picture in Fig. 5 after being rotated at a preset angle.
  • the plant pictures in Fig. 5 and Fig. 6 are divided into rows by the branch lines of the preset direction and the preset row spacing. It can be seen that in Fig. 6
  • the branch line is clearly more appropriate.
  • the preset direction is not limited, as long as it is a fixed direction.
  • the fixed point of the plant picture is used as the rotation center, and the plant picture is rotated according to a fixed preset angle, wherein the fixed point is any point on the plant picture.
  • the center point of the plant image is taken as the rotation center, so as to perform the branching step, which makes the operation more convenient.
  • the number of target tassel in unit row 1-10 are: (0, 0, 0, 1, 0, 0, 0, 0, 0, 2); Preset angle rotation, wherein, the fixed preset angle can be selected as 5°.
  • the step of rotating the plant picture at a preset angle includes relatively rotating the plant picture at a preset angle, that is, by rotating the branch line at a preset angle relative to the plant picture, based on the method , until the rotation angle of the plant picture reaches 180°, including that the relative rotation angle of the plant picture reaches 180°, or the rotation angle of the branch line reaches 180°.
  • the plant pictures are rotated through different methods to improve the applicability.
  • the fixed point of the branch line is used as the rotation center, and the branch line is rotated according to a fixed preset angle.
  • the center point of the branch line is the rotation center, which makes the algorithm simpler.
  • the number of objects in each unit row of the current picture is determined by the method of automatic identification.
  • the target object in the current picture is identified by the target object model, and the number of the target objects is determined, wherein the target object model is learned through deep learning of historical target object pictures in advance.
  • the server side can obtain the corn tassel model through deep learning of historical corn tassel pictures in advance, and then process the unit rows of the plant pictures based on the corn tassel model, and identify the corn tassel (target) of each unit row.
  • Maize tassel counts the number of maize tassel per unit row. In other embodiments, the maize tassel of each unit row in each image is identified and counted by manual methods, but this method is inefficient.
  • manual inspection or random inspection is performed to improve the identification accuracy.
  • step 1.2 the number of objects in each unit row in each picture is obtained to determine the set of object numbers. Based on the foregoing steps, the number of objects in each unit row of each picture is obtained and counted, and a set of the number of objects is obtained, which is convenient for subsequent analysis and processing.
  • each picture and each unit row can be numbered, and each target quantity corresponds to a unique code, for example ( Figure 1, unit row 1), so as to analyze the source and determine the association between the unit row and the plant picture sex.
  • Step 1.3 determine the parent row according to the target quantity set.
  • the unit row corresponding to the number of objects is determined as the parent row; or, the number of objects is sorted according to the size of the number of objects The unit row corresponding to the number of the target objects with the maximum preset number is determined as the parent row.
  • the unit row is determined to be the parent row.
  • the target threshold is determined based on the number of targets in the plant image. Under the premise of the same size of the pictures and the same collection method (at a fixed distance directly above the plants), adjust the target threshold according to the fixed distance. If the fixed distance of the collected pictures is high, the picture contains more crops.
  • the target object threshold is 2 at this time.
  • the target object threshold is 2 at this time.
  • sort each of the unit rows according to the number of objects and determine the unit row corresponding to the number of the objects with the preset number of the largest number of objects as the parent row.
  • the number of all the objects is sorted, and the unit row corresponding to the number of several tassel of the preset number with a larger number is used as the parent row.
  • the unit row with the largest number is taken as the parent row, or the first and second largest numbers are taken as the parent row, so as to improve the detection efficiency and accuracy of the parent row.
  • Step S106 hide the parent row, and determine the plant picture after hiding the parent row.
  • the parent row can be detected quickly and accurately.
  • This embodiment also provides a method for detecting the parent of a target, including the method for processing the parent of the target as described in any of the above, and based on The plant image after hiding the male parent row determines the target of the female parent row. By quickly determining the male parent row, the influence of the tassel of the male parent row is reduced in a hidden way, which facilitates the detection of the female parent target and improves the detection efficiency of the female parent tassel.
  • the parent rows can be checked, there is a situation where there are no tassel or the number of tassel in the identified male parent is less than the tassel threshold, and all the parent rows cannot be accurately determined, so the remaining parent rows cannot be confirmed.
  • the unit rows of are all female parent rows, and the following methods are needed to further identify the tassel of the female parent row in order to determine the removal rate of the female parent target.
  • the embodiment of the present application discloses a method for processing the male parent of a target object and a method for detecting the female parent of a target object. After that, the male parent row will no longer be used for target identification, that is, only the target objects in the remaining rows will be detected and recognized, so as to improve the recognition efficiency and accuracy of the female parent target.
  • the method includes:
  • the picture of the plant is shown in Figure 7, and the white box in Figure 7 is the identified target, that is, the corn tassel.
  • the targets may include female and male targets, ie, female and male tassel.
  • the target is saved as a picture of the target, as shown in the white block diagram in Figure 8, in order to reduce the amount of data download and improve the detection efficiency.
  • the saved target pictures are displayed in a regular arrangement on the detection interface, as shown in Figure 8 (the same target pictures here are only examples of arrangement, and the actual target pictures are different). In one embodiment, all target object pictures are displayed, so as to further improve the detection efficiency.
  • the roles of the female parent object and the male parent object will be exchanged according to different environments, so it is necessary to further screen and confirm the pictures of the objects.
  • the pictures of the female parent object with the female parent object are screened in parallel from the object pictures.
  • the tassel is taken as an example, the female parent characteristics include the female parent tassel color, the female parent tassel shape, the leaf color, etc.
  • the female parent characteristics are not limited to the characteristics of the female parent target itself, but also can include environmental characteristics, Growth characteristics, as long as the female parent tassel can be identified by the female parent characteristics, and at the same time, when identifying the female parent target image, not only one type of female parent characteristics, but also a variety of female parent characteristics can be used. This target image to improve the recognition accuracy.
  • the parent object picture is determined by means of parallel confirmation, and the first operation instruction includes one or more of the following: selecting the parent object picture in parallel, The picture of the parent object is deleted in parallel, and the picture of the parent object or the picture of the parent object is deselected in parallel.
  • the arrangement and display of the detection interface has played a role in promoting the parallel identification of the female parent target, especially the manual parallel identification of the female target has played a great role in promoting, because the target object is captured from the plant picture.
  • Pictures small pictures
  • multiple target pictures are displayed on the detection interface at the same time by arrangement, without downloading a large number of pictures, so that users can intuitively and conveniently
  • Parallel identification and determination of the target map of the female parent eliminates the need to download the pictures one by one and judge the female parent target objects one by one, which improves the detection efficiency and accuracy.
  • the target object pictures are sorted according to image parameters, wherein the image parameters include RGB parameters, HSV parameters, and image texture parameters;
  • the second operation instruction of the picture of the target object is to view, enlarge or reduce the picture of the target object.
  • sorting, filtering, labeling, and deleting target pictures on the detection interface it can help improve the confirmation efficiency of the parent target.
  • batches of male parent objects male parent tassel
  • batches of female parent target objects female parent tassel
  • operations such as sorting, screening, labeling, and deletion of the detection interface can be used alone or in combination.
  • it can be filtered by factors such as color, and at the same time, it can be combined with batch labeling operations to speed up screening and improve detection efficiency.
  • batch labeling operations to speed up screening and improve detection efficiency.
  • you can also restore the previous operation to improve user experience and avoid operation mistakes.
  • it is also possible to perform inverse selection on the already screened pictures to improve applicability and screening speed.
  • the method further includes: obtaining the parent object of the target plot based on the number of pictures of the parent object and the number of plants in the target plot removal rate. That is, the tassel detection of the female parent is realized by extracting or remaining pictures of the female target, and further, based on the number of plants in the collected target plot, the emasculation purity of the plants in the plot can be determined.
  • the removal rate of the female parent target can be quickly obtained, so as to judge whether the emasculation is complete and whether it is necessary to continue emasculation, improve the detection timeliness, make the detection results can be applied on-site, and avoid the detection caused by the long detection time and plant growth. The result is inaccurate.
  • the method for processing a parent of a target object provided by the embodiments of the present application has the same technical features as the method for processing a male parent of a target object provided by the above-mentioned embodiments, so it can also solve the same technical problem and achieve the same technical effect.
  • the computer program product of the male parent, the female parent processing method, the device, and the system of the target object provided by the embodiments of the present application includes a computer-readable storage medium storing program codes, and the instructions included in the program codes can be used to execute the foregoing method implementation.
  • the instructions included in the program codes can be used to execute the foregoing method implementation.
  • Embodiments of the present application further provide an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor.
  • the processor executes the computer program, the method for processing a parent of a target object provided by the foregoing embodiments is implemented. A step of.
  • Embodiments of the present application further provide a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is run by a processor, the steps of the object parent processing method of the foregoing embodiment are executed.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Multimedia (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)

Abstract

一种目标物父本处理方法和目标物母本检测方法，涉及杂交育种的技术领域，包括获得在目标地块中采集的植株图片，所述植株图片中的植株沿行比种植形成父本行和母本行；根据分行检测方式确定所述植株图片中的所述父本行；隐藏所述父本行，确定隐藏所述父本行后的植株图片。

Description

目标物父本处理方法和目标物母本检测方法
本申请要求在2020年11月23日提交中国专利局、申请号为202011325388.4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
本申请涉及杂交育种技术领域，例如涉及一种目标物父本处理方法和目标物母本检测方法。
背景技术
随着科学技术的发展，深度学习算法广泛应用于图像目标检测的领域中，通过训练模型从图像中识别出目标特征，来实现对目标对象进行检测目的。
然而，在一些植株杂交育种场合中，对于雌雄同株的作物来说，多采用行比模式种植，即父本行和母本行间隔种植，在这种场景下通过图像检测的方式检测植株的特征，由于识别的图像往往既包括父本植株又包括母本植株，仅基于深度学习算法无法准确识别出当前存在雄穗特征的是父本植株还是母本植株，进而无法准确检测母本雄穗特征或父本雄穗特征。因此，上述识别方法无法对母本雄穗去除情况进行准确检测。
发明内容
本申请提供一种目标物父本处理方法和目标物母本检测方法，通过隐藏父本行，实现检测母本雄穗的目的，提高检测精度。
第一方面，本申请实施例提供了一种目标物父本处理方法，包括：
获得在目标地块中采集的植株图片，其中，所述植株图片中的植株沿行比种植形成父本行和母本行；
根据分行检测方式确定所述植株图片中的所述父本行；
隐藏所述父本行，确定隐藏所述父本行后的植株图片。
第二方面，本申请实施例还提供一种目标物母本检测方法，包括如上任一项所述的目标物父本处理方法，所述方法还包括：
基于隐藏所述父本行后的植株图片确定母本行的目标物。
附图说明
图1为本申请实施例提供的一种目标物父本处理方法流程图；
图2为本申请实施例提供的一种行比种植的植株示意图；
图3为本申请实施例提供的一种的无人机采集图像示意图；
图4为本申请实施例提供的一种基于植株图像的分行操作示意图；
图5为本申请实施例提供的另一种基于植株图像的分行操作示意图；
图6为本申请实施例提供的一种旋转图像示意图；
图7为本申请实施例提供的一种植株图片示意图；
图8为本申请实施例提供的一种检测界面示意图。
具体实施方式
下面将结合附图对本申请的技术方案进行清楚、完整地描述，所描述的实施例是本申请一部分实施例，而不是全部的实施例。
本实施例适用于杂交育种技术中，下面以玉米育种为例进行说明。玉米杂交制种一般采用“行比模式”，把父、母本行按一定比例交叉种植。即母本行连续种植若干行，父本行再连续种植若干行，之后母本行继续连续种植若干行，按照这样父、母本间隔的方式种植。
杂交育种通过采用去雄处理，即在授粉期间，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，从而增加母本行上玉米籽粒的养分，提高产量。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。
制种公司在培育种子时，对种子纯度的要求达99.7％以上。为了获取高纯度种子，种业公司需要彻底去除玉米母本的雄穗，目前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。一种实施例中，通过人工智能图像识别的办法分析无人机拍摄的去雄后地块的图片来做去雄纯度检测可以极大地节省人力和时间并提高去雄检测效率。但是人工智能图像识别只能有效识别玉米雄穗，缺乏有效方式区分父、母本雄穗，即无法识别出当前具有雄穗的该植株属于父本行还是母本行，进而无法确定母本行雄穗的去除率。
为了检测去雄后母本雄穗的去除情况，在一些实施例中，需要人工对每张图片进行筛选，以确认母本雄穗的去除情况，工作量与图片下载量均较大，效率极为低下；在另一种实施例中，通过人工去田间检查是否去雄彻底，但此种 方式效率更低，作业条件恶劣，人力时间等成本较高。
对于行比种植的玉米育种来说，通过无人机拍摄去雄后地块的图片获得植株图像，然而该植株图像中一般包括父本行和母本行，此时的父本行保留玉米雄穗，母本行已经经过去雄操作，同时母本行的雄穗可能存在未去雄完全、遗漏的情况。通过上述图像识别方法无法快速有效地对母本行去雄情况进行检测。
基于此，本申请实施例提供的一种目标物父本处理方法和目标物母本检测方法，通过隐藏父本行，实现去雄后检测母本行雄穗的目的，提高检测精度。
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目标物父本处理方法进行详细介绍，可应用于服务器端。
图1为本申请实施例提供的一种目标物父本处理方法流程图。
参照图1，该目标物父本处理方法，包括以下步骤：
步骤S102，获得在目标地块中采集的植株图片，其中，所述植株图片中的植株沿行比种植形成父本行和母本行，并且植株图片包含多个，多个植株图片覆盖整个所述目标地块，以提高检测准确度。如图2所示，植株图片中包括父本行和母本行沿行比排列的植株，填充有黑色的为父本行植株，空心的为母本行植株。育种领域的去雄操作是需要去除所有母本行植株上的雄穗，而保留父本行植株上的雄穗，故需要检测母本行雄穗去除率。
需要说明的是，采集的植株图片，可以是几十至上百张或上千张覆盖全部种植地块的图片，或者在全部种植地块的若干个抽样地块采集覆盖全部抽样地块的植株图片，即，为了提高检测效率，所述目标地块为抽样地块，通过在整个地块中抽样选择若干个目标地块，减少了覆盖全部种植地块的植株图片数量，提高检测效率。
在一些实施例中，为了提高识别准确度，每个植株图片通过相同的方法采集，便于后续通过相同的模型识别，提高检测准确度。具体的，植株图片在植株正上方固定距离处拍摄采集，采集图像的姿态和方法的一致性可以保证后续识别的一致性，减小模型训练的难度。需要说明的是，采集到的植株图片无需坐标信息，也可以实现之后的步骤。在一些实施例中，该采集的植株图片中具有坐标信息，可以依据此坐标信息对识别出的母本目标物(母本雄穗)进行单独处理。
步骤S104，根据分行检测方式确定所述植株图片中的所述父本行。
示例性的，图2中在植株行中的方框是相机采集到的植株图片，包含父本行和母本行，其父本行和母本行平行并且平行于图片边界线。在一些实施例中，参考图3，在植株行中的方框是相机采集到的植株图片，父本行、母本行与图片 边界线不平行，父本行、母本行的长度不一致，父本行较短，但是存在较多雄穗，其中，星形代表存在雄穗，空心星形代表存在母本行雄穗，实心星形代表存在父本行雄穗。通过分行检测方式可以适应不同情况的植株图片，以便于有效识别父本行。
为了准确检测父本行，在一些实施例中，步骤S104，包括：
步骤1.1)，通过分行操作将所述植株图片分成多个单元行，确定当前图片每个单元行的目标物数量，持续执行以预设角度旋转所述植株图片后通过分行操作将旋转后的植株图片分成多个单元行，确定当前图片每个单元行的目标物数量，直至所述植株图片的旋转角度达到180°。
其中，基于预设方向和预设行间距确定分行线，以所述分行线对所述植株图片进行分行操作，得到多个单元行。在植株图片上以固定的预设行间距进行分行，如图5所示，该单元行1-10中的目标物雄穗数量分别为：(0，0，1，0，0，0，0，0，1，1)。其中预设行间距为合适的间距，需要根据实际父本行和/或母本行的间距而确定，以提高检测准确度。如果预设行间距太大了，分行的单元行可能存在2行或者不同行植株，造成误检，如图4所示，单元行1-4的目标物雄穗数量分别为：(1，0，0，2)，单元行过大，导致单元行4中存在分别属于两行的植株，容易检测不准确；如果预设行间距太小，无法准确检测行，在父本行和/或母本行种植歪了时，容易检测不到行。
需要说明的是，本申请实施例以两个相邻分行线限定单元行，该单元行中的作物被认定为是同一行的作物。图6为图5中植株图片以预设角度旋转后的图片，通过预设方向和预设行间距的分行线分别对图5和图6中的植株图片进行分行，可以看到，图6中的分行线明显更加合适。其中，预设方向不做限制，只要是固定的方向即可。
在一些可选的实施例中，以所述植株图片的固定点为旋转中心，按照固定预设角度将所述植株图片进行旋转，其中，所述固定点为所述植株图片上的任意点。具体的，以植株图片的中心点为旋转中心，以便执行分行步骤，使得操作更加简便。如图6所示，单元行1-10的目标物雄穗数量分别为：(0，0，0，1，0，0，0，0，0，2)；以植株图片中心点并且以固定预设角度旋转，其中，固定预设角度可选为5°。持续旋转该植株图片并统计每张图片的中每个单元行的目标物数量(雄穗数量)，直至当前的植株图片与最初的植株图片间的旋转角度达到180°，即该植株图片的旋转角度达到180°。
在一些实施例中，以预设角度旋转所述植株图片的步骤，包括以预设角度相对旋转所述植株图片，即，通过将分行线以预设角度相对于植株图片进行旋转，基于该方法，直至所述植株图片的旋转角度达到180°的步骤，包括植株图 片的相对旋转角度达到180°，或者分行线的旋转角度达到180°。如此，通过不同的方法对植株图片进行旋转，提高适用性。此时，以分行线固定点为旋转中心，按照固定预设角度将分行线进行旋转，具体的，分行线的中心点为旋转中心，使算法更加简单。
为了提高效率和适用性，通过自动识别的方法确定当前图片每个单元行的目标物数量。具体的，通过目标物模型识别当前图片中的目标物，并确定所述目标物数量，其中，所述目标物模型是通过预先经过深度学习历史目标物图片习得的。服务器端可以预先通过深度学习历史玉米雄穗图片，获得玉米雄穗模型，基于玉米雄穗模型进而处理植株图片的单元行，识别每个单元行的玉米雄穗(目标物)，根据识别出的玉米雄穗统计每个单元行玉米雄穗数量。在其他实施例中，通过人工方法识别每张图片中每个单元行的玉米雄穗并统计数量，但是这种方法效率低下。可选的，在自动识别每个单元行的目标物后，通过人工方法检查或抽查，以提高识别准确度。
步骤1.2)，获得每张图片中每个单元行的目标物数量确定目标物数量集合。基于前述步骤获得每张图片每个单元行的目标物数量进行统计，得到目标物数量集合，便于后续分析处理。其中，可以对每张图片和每个单元行进行编号，每个目标物数量都对应有唯一的编码，例如(图1，单元行1)，以便于分析溯源，确定单元行和植株图片的关联性。
步骤1.3)，根据所述目标物数量集合确定父本行。
其中，若所述目标物数量大于目标物阈值时，则将所述目标物数量对应的单元行确定为父本行；或者，按照所述目标物数量的大小进行排序，将所述目标物数量最多的预设个数的所述目标物数量对应的单元行确定为父本行。
在一些实施例中，由于父本行植株的雄穗保留，母本行植株的雄穗经过去雄处理，因此可知父本行的雄穗相比于母本行的雄穗的数量较多。当单元行统计得到的雄穗数量(目标物数量)大于雄穗阈值(目标物阈值)时，则认定该单元行是父本行。其中，目标物阈值基于植株图片中目标物的数量进行确定。在图片大小一致的前提下，采集方法一致的情况下(在植株正上方固定距离处)，根据固定距离调整目标物阈值，若采集图片的固定距离较高，此时图片中包含较多的作物，则增加目标物阈值，若采集图片固定距离较低，此时含作物较少，则减小目标物阈值。例如，当固定图片大小、固定距离采集图片的对应的地块实际面积为4*3m时，此时目标物阈值为2。当对应图片中单元行中识别的雄穗大于2时，则认定为父本行。
或者，按照目标物数量将每个所述单元行进行排序，将所述目标物数量最多的预设个数的所述目标物数量对应的单元行确定为父本行。示例性地，将所 有的目标物数量排序，将数量较多的预设个数的几个雄穗数量对应的单元行作为父本行。一般将数量最大的那个单元行作为父本行，或者取数量第一大和第二大的为父本行，以提高父本行检测效率和准确率。
步骤S106，隐藏所述父本行，确定隐藏所述父本行后的植株图片。通过将植株图片中的父本行隐藏的方法，减少了父本雄穗的干扰，直接提高后续母本雄穗检测的准确度，保证母本雄穗检测的精度。
通过上述实施例中的方法，可以快速准确得检测得到父本行，本实施例还提供一种目标物母本检测方法，包括如上任一项所述的目标物父本处理方法，以及，基于隐藏所述父本行后的植株图片确定母本行的目标物。通过快速确定父本行，以隐藏的方式减少父本行的雄穗的影响，便于母本目标物的检测，提高母本雄穗检测效率。在一些实施例中，虽然可以检查一部分父本行，但是存在识别的父本中没有雄穗或雄穗数量不到雄穗阈值的情况，无法准确确定所有的父本行，故无法确认剩下的单元行都是母本行，需要通过下面的方法进一步识别母本行的雄穗，以便确定母本目标物的去除率。
本申请实施例公开了一种目标物父本处理方法和目标物母本检测方法，通过分行检测方式确定植株图片中的父本行，将确定的父本行进行隐藏处理，隐藏了父本行的植株图片，之后将不再对该父本行进行目标物识别，即仅检测识别剩余其他行的目标物，以提高母本目标物的识别效率和准确性。
一种实施例中，该方法包括：
并行识别多个隐藏所述父本行后的植株图片中的目标物，其中，所述父本行上的目标物为父本目标物，所述母本行上的目标物为母本目标物；保存所述植株图片中识别的所述目标物为目标物图片，其中，所述目标物图片包括母本目标物图片、父本目标物图片中的一种或两种；在检测界面中展示所述目标物图片；从所述检测界面中筛选所述母本目标物图片；提取所述母本目标物图片。
其中，植株图片如图7所示，图7中白色方框为识别出的目标物，即玉米雄穗。
为了提高检测效率，通过并行识别目标物的方式，同时处理多个隐藏所述父本行后的植株图片，并且基于目标物模型自动识别目标物，极大加快了检测速度，此时，检测到的目标物可以包括母本目标物和父本目标物，即母本雄穗和父本雄穗。
在通过并行识别方法确定目标物后，保存目标物为目标物图片，如图8中白色框图所示，以便减小数据下载量，提高检测效率。此后，为了提高后续检测筛选效率，在检测界面以规律排列的方式展示保存的目标物图片，如图8所 示(此处相同的目标物图片仅为排布示例，实际目标物图片不同)。一种实施例中，展示所有的目标物图片，以便进一步提高检测效率。
母本目标物和父本目标物的角色会根据不同的环境而调换，故需要对目标物图片进一步筛选确认。为了区分母本目标物和父本目标物，从所述目标物图片中并行筛选具有母本特征的母本目标物图片，其中，母本特征为反应母本目标物的任意特征，以母本雄穗为例，母本特征包括母本雄穗颜色、母本雄穗形状、叶子颜色等等，需要说明的是，母本特征不限于母本目标物本身的特征，也可以包括环境特征、生长特征，只要能通过母本特征识别出母本雄穗即可，同时，识别母本目标物图片时，不仅可以采用一种母本特征，还可采用多种母本特征，综合考虑确定母本目标物图片，以提高识别准确率。
进一步的，响应于用户的第一操作指令，通过并行确认的方式确定所述母本目标物图片，所述第一操作指令包括以下一种或多种：并行选择所述母本目标物图片、并行删除所述父本目标物图片、并行逆选所述母本目标物图片或所述父本目标物图片。存在父母本雄穗差异非常小的情况，通过机器学习无法准确识别，但是通过人工基于母本特征可以从图片中非常快速地区分两者的区别，故通过人工并行识别的方式，快速确定母本目标物，提高母本目标物检测准确度。
需要说明的是，检测界面的排布、展示对并行识别母本目标物起到了促进作用，特别是人工并行识别母本目标物起到了极大的促进作用，由于从植株图片中截图了目标物图片(小图)，剔除了不必要因素，仅留下需要确认的目标物图片，并且多个目标物图片通过排布的方式同时展示于检测界面，无需下载大量图片，使得用户可以直观方便地并行识别确定母本目标图，无需一一下载图片、一一判断母本目标物，提高检测效率和准确度。作为一种可选的实施例，根据图像参数对所述目标物图片进行排序，其中，所述图像参数包括RGB参数、HSV参数和图像纹理参数；在又一种实施例中，响应于用户针对所述目标物图片的第二操作指令，查看、放大或缩小所述目标物图片。通过在检测界面排序、筛选、标注、删除目标物图片，辅助提高母本目标物的确认效率。示例性地，在上述筛选过程中，可以根据作业人员的操作进行批量父本目标物(父本雄穗)的删除，也可以进行批量母本目标物(母本雄穗)的提取。此外，检测界面的排序、筛选、标注、删除等操作可以单独使用也可以多种一起使用，例如，可以通过颜色等因素进行筛选，同时配合批量标注操作，加快筛选速度，提高检测效率。在删除或提取错误时，还可以恢复前一步操作，提高用户体验，避免操作失误。进一步的，还可以对已经筛选的图片进行逆选，提高适用性和筛选速度。
为了提高用户体验，在提取母目标物图片的步骤之后，所述方法还包括：基于母本目标物图片的数量和所述目标地块的植株数量，获得所述目标地块的母本目标物去除率。即通过提取或者剩余的母目标物图片，实现母本雄穗检测，进一步，基于采集到的目标地块的植株数量，可以确定地块中植株的去雄纯度。本实施例中快速获得母本目标物去除率，以便于判断去雄是否彻底，是否需要继续去雄，提高检测时效性，使得检测结果可以现场应用，避免由于检测时间漫长而植株生长造成的检测结果不准确。
本申请实施例提供的目标物母本处理方法，与上述实施例提供的目标物父本处理方法具有相同的技术特征，所以也能解决相同的技术问题，达到相同的技术效果。
本申请实施例所提供的目标物父本、母本处理方法、装置以及系统的计算机程序产品，包括存储了程序代码的计算机可读存储介质，所述程序代码包括的指令可用于执行前面方法实施例中所述的方法，具体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清楚地了解到，为描述的方便和简洁，上述描述的系统和装置的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
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实施例提供的目标物父本处理方法的步骤。
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实施例的目标物父本处理方法的步骤。

Claims (10)

  1. 一种目标物父本处理方法，包括：
    获得在目标地块中采集的植株图片，其中，所述植株图片中的植株沿行比种植形成父本行和母本行；
    根据分行检测方式确定所述植株图片中的所述父本行；
    隐藏所述父本行，确定隐藏所述父本行后的植株图片。
  2. 根据权利要求1所述的目标物父本处理方法，其中，根据分行检测方式确定植株图片中的所述父本行的步骤，包括：
    通过分行操作将所述植株图片分成多个单元行，确定当前图片每个单元行的目标物数量，持续执行以预设角度旋转所述植株图片后通过分行操作将旋转后的植株图片分成多个单元行，确定当前图片每个单元行的目标物数量，直至所述植株图片的旋转角度达到180°；
    获得每张图片中每个单元行的目标物数量确定目标物数量集合；
    根据所述目标物数量集合确定父本行。
  3. 根据权利要求2所述的目标物父本处理方法，其中，根据所有的目标物数量集合确定父本行步骤，包括：
    若所述目标物数量大于目标物阈值时，则将所述目标物数量对应的单元行确定为父本行；
    或者，
    按照所述目标物数量的大小进行排序，将所述目标物数量最多的预设个数的所述目标物数量对应的单元行确定为父本行。
  4. 根据权利要求2所述的目标物父本处理方法，其中，通过分行操作将所述植株图片分成多个单元行的步骤，包括：
    基于预设方向和预设行间距确定分行线，以所述分行线对所述植株图片进行分行操作，得到多个单元行。
  5. 根据权利要求2所述的目标物父本处理方法，其中，确定当前图片每个单元行的目标物数量的步骤，包括：
    通过目标物模型识别当前图片中的目标物，并确定所述目标物数量，其中，所述目标物模型是通过预先经过深度学习历史目标物图片习得的。
  6. 根据权利要求2所述的目标物父本处理方法，其中，以预设角度旋转所述植株图片的步骤，包括：
    以所述植株图片中固定点为旋转中心，按照固定预设角度将所述植株图片进行旋转，其中，所述固定点为所述植株图片上的任意点。
  7. 一种目标物母本检测方法，包括权利要求1～6任意一项所述的目标物父本处理方法，所述方法还包括：
    基于隐藏所述父本行后的植株图片确定母本行的目标物。
  8. 根据权利要求7所述的目标物母本检测方法，其中，基于隐藏了所述父本行的植株图片确定母本行的目标物的步骤，包括：
    并行识别多个隐藏所述父本行后的植株图片中的目标物，其中，所述父本行上的目标物为父本目标物，所述母本行上的目标物为母本目标物；
    保存所述植株图片中识别的所述目标物为目标物图片，其中，所述目标物图片包括母本目标物图片、父本目标物图片中的一种或两种；
    在检测界面中展示所述目标物图片；
    从所述检测界面中并行筛选所述母本目标物图片；
    提取所述母本目标物图片。
  9. 根据权利要求8所述的目标物母本检测方法，其中，从所述检测界面中并行筛选所述母本目标物图片的步骤，包括：
    从所述目标物图片中并行筛选具有母本特征的母本目标物图片；
    响应于用户的第一操作指令，通过并行筛选的方式筛选所述母本目标物图片，所述第一操作指令包括以下一种或多种：并行选择所述母本目标物图片、并行删除所述父本目标物图片、并行逆选所述母本目标物图片或所述父本目标物图片。
  10. 根据权利要求8所述的目标物母本检测方法，其中，在检测界面中展示所述目标物图片的步骤，包括：
    根据图像参数对所述目标物图片进行排序，并按照顺序展示于检测界面，其中，所述图像参数包括RGB参数、HSV参数和图像纹理参数；
    响应于用户针对所述目标物图片的第二操作指令，查看、放大或缩小所述目标物图片。
PCT/CN2020/131745 2020-11-23 2020-11-26 目标物父本处理方法和目标物母本检测方法 WO2022104866A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011325388.4A CN112487906A (zh) 2020-11-23 2020-11-23 目标物父本处理方法和目标物母本检测方法
CN202011325388.4 2020-11-23

Publications (1)

Publication Number Publication Date
WO2022104866A1 true WO2022104866A1 (zh) 2022-05-27

Family

ID=74933152

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/131745 WO2022104866A1 (zh) 2020-11-23 2020-11-26 目标物父本处理方法和目标物母本检测方法

Country Status (2)

Country Link
CN (1) CN112487906A (zh)
WO (1) WO2022104866A1 (zh)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9478038B2 (en) * 2014-03-31 2016-10-25 Regents Of The University Of Minnesota Unsupervised spatio-temporal data mining framework for burned area mapping
CN106373150A (zh) * 2016-08-15 2017-02-01 中国农业大学 一种基于高分辨率遥感数据纹理分析的制种玉米识别方法
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
CN109087312A (zh) * 2018-07-25 2018-12-25 深圳高科新农技术有限公司 一种无人机航线自动规划方法和系统

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9959771B1 (en) * 2015-12-18 2018-05-01 Amazon Technologies, Inc. Unmanned aerial vehicle routing using real-time weather data
CN107945175A (zh) * 2017-12-12 2018-04-20 百度在线网络技术（北京）有限公司 图像的评价方法、装置、服务器及存储介质
CN111062314B (zh) * 2019-12-13 2021-11-02 腾讯科技（深圳）有限公司 图像选取方法、装置、计算机可读存储介质及电子设备

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9478038B2 (en) * 2014-03-31 2016-10-25 Regents Of The University Of Minnesota Unsupervised spatio-temporal data mining framework for burned area mapping
CN106373150A (zh) * 2016-08-15 2017-02-01 中国农业大学 一种基于高分辨率遥感数据纹理分析的制种玉米识别方法
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
CN109087312A (zh) * 2018-07-25 2018-12-25 深圳高科新农技术有限公司 一种无人机航线自动规划方法和系统

Also Published As

Publication number Publication date
CN112487906A (zh) 2021-03-12

Similar Documents

Publication Publication Date Title
CN108647652B (zh) 一种基于图像分类和目标检测的棉花发育期自动识别方法
Fu et al. A novel image processing algorithm to separate linearly clustered kiwifruits
Parvathi et al. Detection of maturity stages of coconuts in complex background using Faster R-CNN model
Yin et al. Computer vision and machine learning applied in the mushroom industry: A critical review
Li et al. In-field cotton detection via region-based semantic image segmentation
Sun et al. Image processing algorithms for infield single cotton boll counting and yield prediction
Zhou et al. Using colour features of cv.‘Gala’apple fruits in an orchard in image processing to predict yield
US9008409B2 (en) Method and system for digital image analysis of ear traits
CN111709489A (zh) 一种基于改进YOLOv4的柑橘识别方法
CN109447945B (zh) 基于机器视觉和图形处理的小麦基本苗快速计数方法
CN110796011B (zh) 一种基于深度学习的稻穗识别方法、系统、装置及介质
Rasti et al. A survey of high resolution image processing techniques for cereal crop growth monitoring
WO2021208407A1 (zh) 目标物检测方法、装置和图像采集方法、装置
CN112990103A (zh) 一种基于机器视觉的串采二次定位方法
CN111462058B (zh) 一种水稻有效穗快速检测方法
CN113191334B (zh) 一种基于改进CenterNet的植物冠层密集叶片计数方法
Sunoj et al. Sunflower floral dimension measurements using digital image processing
CN109522899A (zh) 咖啡成熟果实的检测方法、装置及电子设备
CN111784764A (zh) 一种茶叶嫩芽识别与定位算法
CN114818909A (zh) 一种基于作物长势特征的杂草检测方法和装置
WO2022104867A1 (zh) 目标物的特征检测方法和装置
Li et al. An automatic approach for detecting seedlings per hill of machine-transplanted hybrid rice utilizing machine vision
CN115661650A (zh) 一种基于物联网数据监测的农场管理系统
CN113313692B (zh) 一种基于航拍可见光图像的香蕉幼株自动识别与计数方法
Gao et al. Maize seedling information extraction from UAV images based on semi-automatic sample generation and Mask R-CNN model

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: 20962157

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: 20962157

Country of ref document: EP

Kind code of ref document: A1