WO2022104867A1

WO2022104867A1 - Feature detection method and device for target object

Info

Publication number: WO2022104867A1
Application number: PCT/CN2020/131746
Authority: WO
Inventors: 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: 2020-11-23
Filing date: 2020-11-26
Publication date: 2022-05-27
Also published as: CN112381028A

Abstract

The present disclosure relates to the technical field of image feature detection. Disclosed are a feature detection method and device for a target object. The method comprises: obtaining target object pictures acquired in a target area; processing the multiple target object pictures in parallel; recognizing first features in the target object pictures; storing the first features, recognized in the target object pictures, as first feature pictures; displaying multiple first feature pictures in a detection interface; screening, in parallel from the detection interface, second feature pictures, which have second features, in the first feature pictures; and extracting the screened second feature pictures.

Description

Target feature detection method and device

This application claims the priority of the Chinese Patent Application No. 202011325389.9 filed with the China Patent Office on November 23, 2020, the entire contents of which are incorporated herein by reference.

technical field

The present application relates to the technical field of image object detection, for example, to a method and device for feature detection of objects.

Background technique

With the development of science and technology, deep learning algorithms are widely used in the field of image target detection. The purpose of detecting corresponding target objects is achieved by training models to identify target features from images.

However, in some practical applications, first, the image features of the target object may not be fixed. For example, for monoecious crops, the tassel may be the male parent or the female parent. Based on deep learning algorithms Only the presence of the tassel can be identified, but it cannot be accurately identified whether the male parent plant or the female parent plant currently has a tassel. In addition, the image used for identification is not conducive to the accurate identification of image features of the algorithm due to the acquisition environment factors such as light and mutual occlusion of target objects. Based on the foregoing situation, its features cannot be accurately detected. Currently, it is generally used manually to detect features directly in the field, which has large workload, low efficiency and high cost.

SUMMARY OF THE INVENTION

The present application provides a feature detection method and device for a target object. By identifying a first feature picture in parallel, and identifying a second feature in the first feature picture through parallel cascading, the feature detection of the target object is realized and the detection accuracy is improved. and detection efficiency.

In a first aspect, an embodiment of the present application provides a method for detecting features of a target, including:

Obtain the image of the object collected in the target area;

processing a plurality of the target object pictures in parallel, and identifying the first feature in the target object pictures;

Save the first feature identified in the target object picture as a first feature picture;

displaying a plurality of the first feature pictures in the detection interface;

Screen the second feature pictures with the second feature in the first feature pictures in parallel from the detection interface;

Extracting the filtered second feature picture.

In a second aspect, an embodiment of the present application also provides a feature detection device for a target, including:

The acquisition module obtains the image of the target object collected in the target area;

an identification module, which processes a plurality of the target object pictures in parallel, and identifies the first feature in the target object pictures;

a saving module, saving the first feature identified in the target object picture as a first feature picture;

a display module, displaying a plurality of the first characteristic pictures in the detection interface;

A screening module, which screens the second feature picture with the second feature in the first feature picture in parallel from the detection interface;

The extraction module extracts the screened second feature picture.

Description of drawings

1 is a flowchart of a method for detecting a feature of a target according to an embodiment of the present application;

Fig. 2 is the schematic diagram of the parent plant of a kind of Gypsophila planting pattern provided by the embodiment of the application;

3 is a schematic diagram of a picture of a target object with a first feature identified according to an embodiment of the present application;

FIG. 4 is a schematic diagram of a first feature picture provided by an embodiment of the present application;

5 is a schematic diagram of a detection interface provided by an embodiment of the present application;

FIG. 6 is a functional block diagram of a feature detection apparatus for a target object provided by an embodiment of the present application.

Detailed ways

The technical solutions of the present application will be clearly and completely described below with reference to the accompanying drawings, and the described embodiments are part of the embodiments of the present application, but not all of the embodiments.

In the actual production process, there is a bottleneck through the deep learning method, and it is impossible to independently identify the fine features of a certain feature, thereby affecting the recognition accuracy and efficiency. Based on this, the embodiment of the present application provides a feature of a target object The detection method and device can quickly and effectively detect the indistinguishable features of the target through the method of cascading and parallel identification, and improve the detection accuracy and detection efficiency.

In order to facilitate the understanding of this embodiment, a method for detecting a feature of a target object disclosed in the embodiment of this application is first introduced in detail, which can be applied to the server side.

FIG. 1 is a flowchart of a method for detecting a feature of a target according to an embodiment of the present application.

Referring to FIG. 1 , an embodiment of the present application provides a feature detection method for a target object, including:

In step S102, a picture of the target object collected in the target area is obtained.

Step S104: Process a plurality of the target object pictures in parallel to identify the first feature in the target object pictures.

Step S106, saving the first feature identified in the target object picture as a first feature picture.

Step S108, displaying a plurality of the first characteristic pictures in the detection interface.

Step S110: Screen second feature pictures with second features in the first feature pictures in parallel from the detection interface.

Step S112, extracting the filtered second feature picture.

Through the above method, in the target area, the first feature picture is obtained through parallel autonomous identification, and then the second feature picture is obtained by parallel screening based on the first feature picture. High-accuracy detection results, to achieve the purpose of fast and efficient detection.

In one embodiment, the first feature is a corn tassel, the second feature is one or more of tassel shape, tassel color, and leaf color, and the first feature picture is a corn tassel Picture, the second characteristic picture is a picture of the female parent tassel. The picture of the target object is a picture of corn, and the cascading identification method in this embodiment determines the picture of the female parent tassel from the corn picture, which solves the problem that the female parent tassel cannot be quickly and accurately detected.

In cross-breeding, tasseling is often used, that is, during pollination, the tassel of the female parent is removed, leaving only the tassel of the male parent. During pollination, the fruit on the plant of the female parent is the pollen of the male parent. It is combined with the eggs of the female parent to achieve hybrid seed production. If the female parent is not emasculated, 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.

Maize hybrid seed production generally adopts the row ratio mode, that is, the parent row is cross-planted in a certain proportion. Because the male parent row does not produce seeds in this row ratio mode, resulting in waste of land, people often replace the male parent row with the female parent row in whole or in part, and plant some male parent rows sporadically in the middle of the female parent row to provide pollen, commonly known as "" Gypsophila" planting mode. The Gypsophila model can often increase the yield by more than 30% compared to the row-by-row model due to the higher proportion of female parents. As shown in Figure 2, the circle in Figure 2 is the female parent plant, and the star is the male parent plant. All female parent plants need to be removed from the tassel during operation.

When seed production companies are cultivating seeds, the requirements for the purity of seeds are over 99.7%, and the requirements for the accuracy of emasculation detection are also increased. 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, the purity detection of emasculation is performed by analyzing the pictures of the plots after emasculation taken by the drone by the method of artificial intelligence image recognition, which can greatly save manpower and time, and improve the efficiency of emasculation detection. However, artificial intelligence image recognition can only effectively identify corn tassel, and there is no effective way to distinguish the parental tassel, that is, it cannot specifically identify whether the current plant with tassel belongs to the male parent or the female parent, and thus cannot achieve accurate female parent tassel. detection.

The applicant's research found that, especially in the Gypsophila planting method, because the parents are very close together, and the parents of corn are planted closely and staggered, it is more difficult to distinguish the tassel of the parents by image recognition methods. One solution is to manually go to the field to check whether the emasculation is complete, which is inefficient, has poor operating conditions, and is time-consuming, labor-intensive and expensive. For pictures, the machine-recognized pictures are manually confirmed, and the screening of female parent tassel by confirming each picture requires a lot of work, large downloads of pictures, and extremely low efficiency.

Based on this, through the target detection method in this embodiment, the first feature picture (maize tassel) is automatically identified in parallel, and the method for obtaining the target feature (female tassel) corresponding to the second feature picture is further manually filtered in parallel, The rapid detection of the target feature (the female parent tassel) is realized, and it can be applied to the field of emasculation to obtain results quickly, thereby assisting emasculation.

In some embodiments, the first feature is a corn bud, the second feature is one or more of bud shape, bud color, and leaf color, the first feature picture is a bud picture, and the second feature picture is a female parent bud picture , through the above method, the picture of the female parent flower bud can be quickly and effectively detected from the corn picture.

In another embodiment, the first feature is a foreign object on a power line, the second feature is one or more of a split pin, an anti-vibration hammer, and a wire clip, the first feature picture is a foreign object picture, and the The second characteristic picture is a picture of a split pin, a picture of a shock-proof hammer, or a picture of a wire clip. This embodiment can also be applied to the field of electric power inspection. Through parallel foreign object detection and further cascading screening to determine a picture of a split pin, a picture of an anti-vibration hammer, or a picture of a wire clip, the purpose of quickly detecting specific foreign objects is achieved.

It should be noted that the target object picture includes a corn picture and a power line picture, and the corresponding first feature and second feature change correspondingly with the change of the target object picture, which is not limited here.

In the practical application embodiment, upload all the target object pictures collected in the sampling area, identify the first feature in parallel, save the first feature picture with the first feature, and display it on the detection interface, based on the operation The instruction selects the second feature picture from the detection interface, extracts the second feature picture, obtains the first feature picture through parallel recognition, and confirms the second feature picture in parallel, which further improves the detection efficiency and makes up for the uncertainty of machine vision detection. , so that the detection accuracy can reach more than 99%.

In some embodiments, step S102 further includes the following steps:

The collected pictures of the target objects are uploaded in batches, and the pictures of the target objects are collected in the target area by the same method.

Among them, the target area is the sampling area, and multiple target areas are randomly sampled in a piece of land, and dozens to hundreds of target object pictures covering the target area are collected to adapt to the large land area and improve the detection efficiency. Or, if the target area is the entire land parcel, at this time, it is necessary to collect the target object pictures covering all the land parcels. The number of pictures is large, which will reduce the detection efficiency.

In order to improve the accuracy of subsequent image recognition, the method for collecting the image of the target object is the same. In one embodiment, the image of the target object is collected at a fixed distance directly above the target object, and the consistency of the posture of the collected image can ensure the subsequent recognition of the image. consistency. Specifically, the UAV receives the acquisition instruction and controls the UAV to hover over each target in the same acquisition attitude, and collect multiple pictures of the target. It should be noted that the collected image of the target object does not require coordinate information, and the subsequent steps can also be implemented. In some embodiments, the collected image of the target object has coordinate information, and the detection result in the detected second feature image can be separately processed according to the coordinate information.

After collecting the target object pictures, upload the collected target object pictures in batches, or upload the collected target object pictures in real time when collecting the target object pictures. The former reduces the communication requirements and improves the detection efficiency, while the latter requires communication. higher. In one embodiment, after the drone collects multiple target areas, it obtains multiple target images, and transmits them to the server through batch uploading (tens, hundreds, or thousands of images can be uploaded together).

In order to improve the detection accuracy and detection efficiency, in step S104: identify the first features in the plurality of target object pictures in parallel by using the first feature model, the first feature model is pre-learned by the deep learning history of the first feature picture Got it.

In this embodiment, by combining cloud computing and AI, the first feature model is preselected through deep learning history to obtain the first feature model, and based on the first feature model, the uploaded target image is processed with high concurrency, and the first feature is quickly identified and marked. out the first feature. Applied to the detection of maize tassel, the time for identifying the parent tassel of Gypsophila has been greatly shortened, and the real-time detection of the effect of Gypsophila emasculation is realized. The server side can obtain the corn tassel model through deep learning of historical corn tassel pictures in advance, and then process all uploaded corn pictures (target images) in high concurrency based on the corn tassel model, process each uploaded corn picture in parallel, and identify the corn. Tassel (first feature). It should be noted that the corn tassel model at this time is used to identify the corn tassel, and the identified corn tassel picture includes the male parent tassel picture and the female parent tassel picture (second feature picture), which needs to be further screened. OK to get a picture of the female tassel.

It can be understood that through the method of deep learning and the method of recognizing multiple target object pictures through parallel processing, in one embodiment, all the target object pictures can be processed at the same time, the first feature in the target object picture can be quickly determined, and the detection rate can be improved. In addition, because the acquisition method of the target image is consistent, the accuracy of the recognition feature is greatly improved, and the detection accuracy is improved.

In step S106, after the first features are identified in parallel, the first features are marked in parallel, and the marked first feature pictures are stored in parallel as the first feature pictures. Wherein, the marking can be realized by any marking, as long as the first feature can be seen conspicuously and clearly, for example, it can be a circular frame, a rectangular frame, a polygonal frame and so on. In this method, a plurality of first feature pictures are obtained through the methods of parallel identification, marking and saving, thereby improving the detection efficiency.

In one embodiment, the target object picture is a corn picture including corn tassel features, as shown in FIG. 3 , and the white box marks the identified first feature. The first feature picture is a small picture of the first feature identified from the target object picture, as shown in Figure 4, that is, each white box area in Figure 3 is extracted from the large picture (target object picture). Figure (first feature picture). By saving the first feature picture, the download volume of subsequent screening is facilitated, the detection efficiency is improved, and it is convenient to further identify and screen the second feature from the first feature picture, avoiding the step of directly screening the second feature from the target object picture, greatly improving the detection efficiency. Improved detection accuracy.

In some embodiments, in step S108, the plurality of first feature pictures are sorted according to image parameters, and displayed on the detection interface simultaneously in order, wherein the image parameters include RGB parameters, HSV parameters and image texture parameters.

The detection interface is shown in FIG. 5 , and the saved first feature pictures are arranged and displayed on the interface, so that a plurality of first feature pictures can be displayed on one page. If the number of the first feature pictures is too large and exceeds the scope of the interface, or the first feature pictures are too small to be accurately screened, or the screening efficiency is low when the first feature pictures are arranged in a small number, you can, according to the operation instructions input by the user, The arrangement of the first feature pictures is adaptively adjusted to improve the screening experience. For example, in some embodiments, in response to the second operation instruction for the first feature picture, select and view the target object picture where the first feature picture is located, that is, click on the thumbnail image, or view the image where the thumbnail image is located. A large picture, so that the first feature picture that is difficult to confirm can still be confirmed by opening its linked large picture. For another example, in response to the third operation instruction for the first feature picture, the first feature picture is enlarged or reduced, and each first feature picture is further confirmed, and further, all the first feature pictures can be further confirmed. The picture is enlarged or reduced so that the feature pictures can be arranged adaptively, the first feature picture can be adjusted to an appropriate size according to the different screening environment, and the number of the first feature pictures displayed on the detection interface can be adjusted, which is convenient for subsequent operations and improves the detection efficiency. . In other embodiments, a drag operation can also be used to display the first feature pictures beyond the interface range, so as to ensure that all the first feature pictures undergo subsequent screening.

It should be noted that the first feature pictures are displayed on the detection interface in a side-by-side arrangement, which is convenient for synchronous detection of multiple feature pictures, as shown in FIG. 5 , the content of the first feature picture is only an example, and the first Feature pictures vary. Further, by arranging multiple rows side by side, the regularity of the arrangement of the first feature pictures is improved, which facilitates the subsequent screening according to the rules, improves the user experience, and further improves the detection efficiency. In one embodiment, all the first feature pictures are displayed on the detection interface, and all the first feature pictures can be screened at the same time, which further improves the detection efficiency.

In order to improve the detection accuracy and detection efficiency of the feature, step S110 includes: in response to the first operation instruction, determining a second feature picture with the second feature in the first feature picture by means of parallel identification, wherein the The operation instructions include one or more of the following: manual parallel selection of the second feature picture, manual parallel deletion of the second feature picture, manual parallel inverse selection of the second feature picture.

By means of manual screening, the second feature picture is directly screened from the detection interface, which can increase the detection efficiency. It should be noted that the manual screening user knows the second feature corresponding to the second feature picture of the target area, and then passes Quickly identify the second feature and quickly determine the second feature picture. The first feature picture set is displayed on the detection interface, and a plurality of first feature pictures are processed in parallel by manual screening, and a plurality of second feature pictures are screened at the same time, so as to improve the detection efficiency. It should be noted that this embodiment can be generally used for feature detection and confirmation in a combination of high-throughput AI and artificially assisted recognition, and can be applied to any scenario that requires manual confirmation and fine-tuning of AI recognition results.

Specifically, enter the detection interface, as shown in Figure 5 as a partial interface, the first feature pictures are regularly arranged on the detection interface, and the second feature pictures are directly selected by manual screening and sorting. Pictures other than the second feature picture are deleted. In the application of corn tassel, when the corn tassel pictures are regularly arranged on the detection interface, the color, shape, and color of the tassel are manually screened to determine whether the picture of the corn tassel is the female parent tassel. If it is the female parent tassel, it is determined as the female parent tassel picture, otherwise it can be deleted. Since there is no fixed female parent tassel, that is, the role of the parent is not fixed, the machine learning method cannot identify the unfixed female parent tassel, and the operator can quickly identify the maize tassel based on the known parental tassel and the field environment. The ear picture identifies the female parent tassel in this target area/plot, which enables screening.

It should be noted that the second feature includes not only the own features of the first feature, but also the environmental features, growth features, etc. of the first feature. The second feature picture is determined by fusion of the two features, and the determination method is not limited, and the second feature is not limited, as long as the second feature picture can be determined by the second feature. The first feature picture includes at least a second feature picture, and the second feature picture is determined from the first feature picture, thereby determining the target feature corresponding to the second feature picture, so as to achieve the purpose of detecting the target feature from the target object picture. In the field of corn tasseling, the second feature picture (the female parent tassel picture) is determined from the target object picture (the corn picture), and the target feature of the second feature picture is the female parent tassel, which is determined by the second feature (tassel color , shape, leaf color, etc.)

In the embodiment of the present application, the step of screening the second feature picture can be assisted by any combination of one or more of the functions of sorting, screening, labeling, and deletion, so as to improve the confirmation efficiency of the second feature picture, improve the adaptability of the detection method, and further Improve user experience. Exemplarily, in the above screening process, the male parent tassel can be deleted in batches according to the operation of the operator, and the female parent tassel can also be extracted in batches. When it cannot be determined, an operation of individual annotation or batch annotation can be performed for the first feature picture. Screening by factors such as color can speed up the screening. By sorting the first feature pictures on the detection interface, for example, sorting by color, combined with batch deletion and extraction, the detection efficiency of the second feature pictures can be further improved. When deleting or extracting errors, you can also restore the previous operation, so as to improve the user experience and avoid operation mistakes. At the same time, it is also possible to perform inverse selection on the already screened pictures, so as to improve the diversification and applicability of the screening, and facilitate the improvement of the screening speed.

As an optional embodiment, in order to improve the applicability of detection, the method further includes the following steps:

Based on the number of the second feature pictures and the number of objects in the target area, a target feature removal rate corresponding to the second feature picture in the target area is obtained.

Here, accurate removal rate calculation or verification can be implemented based on the embodiments of the present application, that is, the second feature picture can be quickly determined through the real-time upload of the target object picture, so as to realize the detection of the second feature picture. Further, based on the number of plants in the collected image of the target object, the target feature removal rate in the target area and the plot to which the target area belongs can be determined.

In this embodiment, the number of target object pictures in the target area is not limited, and may range from single digits, dozens, hundreds, or thousands. First, all the corn tassel is detected by the deep learning detection algorithm, then all the corn tassel content is saved in the new first feature picture, and all the first feature pictures are displayed on the detection interface, and the system responds to the job user The first feature image is subjected to a series of corresponding processing operations, wherein the operations include sorting (for example, according to each color channel of RGB/HSV), selection, deletion, annotation, inverse selection, etc. Rapid extraction of non-target features (male parent tassel) and rapid deletion of non-target features (male parent tassel), so as to achieve the purpose of accurate detection of female parent tassel.

In addition, since all the artificially assisted screening process takes place on the first feature picture, the confirmation work of all target object pictures can be completed synchronously (a target object picture may have multiple first feature pictures), thereby achieving ten percent improvement in detection efficiency. times or even hundreds of times.

In some application scenarios, especially when detecting tassel in corn breeding, the requirement for detection time is relatively high. Since the plant is in the process of continuous growth, if the detection cannot be obtained in time (the time required for general image detection is counted in days) ), will not reflect the real situation on site. In this embodiment, the first feature picture set is obtained by collecting the target object pictures in real time, identifying the first feature in parallel, and presenting the first feature picture set in batches to avoid downloading a large amount of data, and at the same time, it is convenient to screen and identify the second feature picture, and the detection result can be obtained in time. The detection time is saved to the minute level, which greatly improves the detection efficiency. This embodiment obtains the detection result of the on-site situation in time, and can be truly applied to application scenarios such as planting and breeding.

This embodiment is applied to the occasions where it is difficult to identify and distinguish the fine features of the image by the machine training method, such as the identification of the parental tassel in the field of cross-breeding, the identification of the female and male buds in the identification of corn buds, the identification of the parental buds in the soybean cross-breeding For example, in the field of drone inspection, it is used to detect characteristic parts such as screws, cotter pins, and wire clips from images.

Further, as shown in FIG. 6 , an embodiment of the present application further provides a feature detection device 600 of a target object, including:

The acquisition module 601 obtains the target object pictures collected in the target area; the identification module 602 processes a plurality of the target object pictures in parallel to identify the first feature in the target object pictures; the saving module 603 saves the target object The first feature identified in the picture is a first feature picture; the display module 604 displays a plurality of the first feature pictures in the detection interface; the screening module 605 filters the first features from the detection interface in parallel A second feature picture with the second feature in the picture; the extraction module 606 extracts the screened second feature picture.

The embodiment of the present application discloses a method and a device for detecting a feature of a target object, uploading a picture of the target object collected in a target area, identifying a first feature in parallel, saving the picture with the first feature, and displaying it on a detection interface. It is shown that the second feature picture is screened and determined from the detection interface based on the operation instruction, and the second feature picture is extracted, and the purpose of accurately detecting the second feature picture of the target object is achieved through cascade recognition, which greatly improves the detection efficiency.

The target feature detection device provided by the embodiment of the present application has the same technical features as the target feature detection method provided by the above embodiments, so it can also solve the same technical problem and achieve the same technical effect.

The computer program product of the method and device for detecting the feature of a target 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 methods described in the foregoing method embodiments. For the specific implementation, reference may be made to the method embodiments, which will not be repeated here.

Those skilled in the art can clearly understand that, for the convenience and brevity of description, for the specific working process of the system and device described above, reference may be made to the corresponding process in the foregoing method embodiments, which will not be repeated here.

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. When the processor executes the computer program, the method for detecting a feature of a target 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 method for detecting a feature of a target object of the foregoing embodiments are executed.

Claims

A feature detection method for a target, comprising:

Obtain the image of the object collected in the target area;

processing a plurality of the target object pictures in parallel, and identifying the first feature in the target object pictures;

Save the first feature identified in the target object picture as a first feature picture;

displaying a plurality of the first feature pictures in the detection interface;

Screen the second feature pictures with the second feature in the first feature pictures in parallel from the detection interface;

Extracting the filtered second feature picture.
The feature detection method of the target object according to claim 1, wherein the step of obtaining the target object picture collected in the target area comprises:

Upload the collected pictures of the target objects in batches, wherein the pictures of the target objects are collected in the target area by the same method, and the target area is a sampling area.
The feature detection method of the target object according to claim 1, wherein the step of processing a plurality of the target object pictures in parallel, and recognizing the first feature in the target object pictures, comprises:

The first features in the plurality of target object pictures are identified in parallel by using a first feature model, and the first feature model is learned in advance through the deep learning history of the first feature pictures.
The feature detection method of the target object according to claim 1, wherein the step of screening the second feature picture with the second feature in the first feature picture from the detection interface in parallel comprises:

In response to the first operation instruction, a second feature picture with the second feature in the first feature picture is determined by means of parallel identification, wherein the operation instruction includes one or more of the following: manually selecting the first feature picture in parallel. Two feature pictures, manual parallel deletion of the second feature picture, manual parallel inverse selection of the second feature picture.
The feature detection method of the target object according to claim 1, wherein the step of displaying a plurality of the first feature pictures in the detection interface comprises:

The plurality of first feature pictures are sorted according to image parameters, and displayed simultaneously on the detection interface in order, wherein the image parameters include RGB parameters, HSV parameters and image texture parameters.
The feature detection method of the target object according to claim 1, wherein the step of displaying a plurality of the first feature pictures in the detection interface comprises:

In response to the second operation instruction for the first feature picture, the target object picture where the first feature picture is located is selected and viewed.
The feature detection method of the target object according to claim 1, wherein the step of displaying a plurality of the first feature pictures in the detection interface comprises:

In response to a third operation instruction for the first feature picture, the first feature picture is enlarged or reduced.
The feature detection method of the target object according to claim 1, further comprising:

Based on the number of the second feature pictures and the number of objects in the target area, a second feature removal rate of the target area is obtained.
The feature detection method of a target object according to claim 1, wherein the first feature is a corn tassel, and the second feature is one or more of a tassel shape, a tassel color, and a leaf color; Alternatively, the first feature is a foreign body on a power line, and the second feature is one or more of cotter pins, anti-vibration hammers, and wire clips; or, the first feature is popcorn buds, and the second feature is One or more of bud shape, bud color, and leaf color.
A feature detection device for a target, comprising:

The acquisition module obtains the image of the target object collected in the target area;

an identification module, which processes a plurality of the target object pictures in parallel, and identifies the first feature in the target object pictures;

a saving module, saving the first feature identified in the target object picture as a first feature picture;

a display module, displaying a plurality of the first characteristic pictures in the detection interface;

A screening module, which screens the second feature picture with the second feature in the first feature picture in parallel from the detection interface;

The extraction module extracts the screened second feature picture.