CN117576560A

CN117576560A - Method, device, equipment and medium for identifying field weeds of northern spring corns

Info

Publication number: CN117576560A
Application number: CN202311540642.6A
Authority: CN
Inventors: 车晓曦; 冉得才; 单奕; 赵寻; 李金伟; 苏靖杰
Original assignee: Sinochem Agriculture Holdings
Current assignee: Sinochem Agriculture Holdings
Priority date: 2023-11-17
Filing date: 2023-11-17
Publication date: 2024-02-20

Abstract

The invention provides a field weed identification method, device, equipment and medium for northern spring corns, and relates to the technical field of image processing. The method comprises the following steps: acquiring a field image of a northern spring corn planting area; inputting the field image into a field weed identification model to obtain a field weed identification result output by the field weed identification model; the first dimension image for training comprises images corresponding to different growth stages of corn; the second dimension image for training comprises an image corresponding to straw coverage and an image corresponding to no straw coverage; the third-dimension image for training comprises images corresponding to different corn cultivation ridge types; the fourth dimension image for training comprises images corresponding to different soil types; the fifth dimension image for training comprises images corresponding to different weeding treatment modes; the sixth-dimension image for training includes images corresponding to different degrees of grass. The invention can improve the training effect of the field weed identification model.

Description

Method, device, equipment and medium for identifying field weeds of northern spring corns

Technical Field

The invention relates to the technical field of image processing, in particular to a field weed identification method, device, equipment and medium for northern spring corns.

Background

In the planting process of northern spring corns, farmers commonly adopt chemical plant protection and weeding twice, and once is a closed weeding operation performed before emergence after sowing; and the other time is the stem leaf weeding operation performed when 3-5 leaves are spread after emergence. The field grass condition can be well controlled through twice chemical plant protection weeding, however, the plant protection operation mode has the phenomenon that pesticides are used excessively, so that the pollution to the field environment is caused, the plant protection cost is increased, and the planting cost is increased.

Even if the chemical adopted for the first field weeding of the corn is a mixed closed chemical, most of weeds in the field are killed after the seedling is in a good weather state, so that the weed base number in the corn field is lower, and only one chemical plant protection weeding is adopted. However, prior to corn ridge sealing, field weeds compete with corn for nutrient nutrition and provide habitat for pests, so that plant protection operation on post-emergence weeds cannot be reduced.

At present, considering that the number of the post-emergence weeds is lower, the field non-differential spraying operation is adopted, the pesticide waste and the environmental pollution are caused, and the planting cost of farmers is increased, so that in the post-emergence weeding operation of the northern spring corn, whether the pesticide is sprayed or not is determined by identifying whether the field images have weeds or not, and the variable plant protection operation can be realized. However, the current weed identification accuracy based on field images is not high, so that the accuracy of variable plant protection operation is reduced, the plant protection effect is finally affected, and the yield and quality of northern spring corns are reduced.

Disclosure of Invention

The invention provides a field weed identification method, device, equipment and medium for northern spring corns, which are used for solving the defect of low field image-based weed identification accuracy in the prior art and realizing high-accuracy field weed identification.

The invention provides a field weed identification method for northern spring corns, which comprises the following steps:

acquiring a field image of a northern spring corn planting area;

inputting the field image into a field weed identification model to obtain a field weed identification result output by the field weed identification model;

the field weed identification model is obtained by training based on a sample field image and a field weed identification result label corresponding to the sample field image;

The sample field image comprises a first dimension image, a second dimension image, a third dimension image, a fourth dimension image, a fifth dimension image and a sixth dimension image; the first dimension image comprises images corresponding to different growth stages of corn; the second dimension image comprises an image corresponding to straw coverage and an image corresponding to no straw coverage; the third dimensional image comprises images corresponding to different corn cultivation ridge types; the fourth-dimension image comprises images corresponding to different soil types; the fifth dimension image comprises images corresponding to different weeding treatment modes; the sixth-dimension image comprises images corresponding to different grass conditions.

According to the field weed identification method for northern spring corns, the first dimension image comprises 2 images corresponding to unfolding leaves, 3 images corresponding to unfolding leaves, 4 images corresponding to unfolding leaves, 5 images corresponding to unfolding leaves and 6 images corresponding to unfolding leaves.

According to the field weed identification method for northern spring corns, the corresponding images of straw coverage comprise images corresponding to irregularly-stacked straws and images corresponding to straws regularly stacked on two sides of a sowing row;

The images corresponding to the irregularly stacked straws comprise a first irregular image and a second irregular image; the density of the straw in the acquisition area of the first irregular image is smaller than the preset density, and the density of the straw in the acquisition area of the second irregular image is larger than or equal to the preset density.

According to the field weed identification method for northern spring corns, the third dimensional image comprises an image corresponding to ridgeless flat planting and an image corresponding to ridging cultivation.

According to the field weed identification method for northern spring corns, the corresponding images for ridging cultivation comprise images corresponding to 2 sowing lines on a 110 cm wide ridge, images corresponding to 1 sowing line on a 65 cm wide ridge and images corresponding to 1 sowing line on a 75 cm wide ridge.

According to the field weed identification method for northern spring corns, the fourth-dimension image comprises an image corresponding to sand and wind soil, an image corresponding to black lime soil, an image corresponding to dark brown soil, an image corresponding to white serosity soil and an image corresponding to meadow soil.

According to the field weed identification method for northern spring corns, the fifth dimension image comprises an image corresponding to a pre-seedling closed weeding treatment mode only, an image corresponding to a post-seedling stem leaf weeding treatment mode only without pre-seedling closed weeding treatment, and an image corresponding to a comprehensive treatment mode with pre-seedling closed weeding treatment and post-seedling stem leaf weeding treatment.

According to the field weed identification method for northern spring corns, the sixth-dimension image comprises an image corresponding to the first grass condition degree, an image corresponding to the second grass condition degree, an image corresponding to the third grass condition degree and an image corresponding to the fourth grass condition degree;

the first grass plot degree is less than the second grass plot degree, the second grass plot degree is less than the third grass plot degree, and the third grass plot degree is less than the fourth grass plot degree.

According to the field weed identification method for northern spring corns, the number of images of the first dimension image, the number of images of the second dimension image, the number of images of the third dimension image, the number of images of the fourth dimension image, the number of images of the fifth dimension image and the number of images of the sixth dimension image are 500 to 1000.

The invention also provides a field weed identification device for northern spring corns, which comprises:

the acquiring module is used for acquiring field images of northern spring corn planting areas;

the identification module is used for inputting the field image into a field weed identification model to obtain a field weed identification result output by the field weed identification model;

And the execution module is used for executing the operation signal of the control equipment on the field weed identification result.

The invention also provides an electronic device comprising a memory, a processor and a computer program stored on the memory and operable on the processor, wherein the processor implements the field weed identification method for northern spring corn as described in any of the above when executing the program.

The present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a field weed identification method for northern spring corn as described in any of the above.

The field weed identification method, the device, the equipment and the medium for the northern spring corn, provided by the invention, input the field image of the northern spring corn planting area into the field weed identification model to obtain the field weed identification result output by the field weed identification model, wherein the field weed identification model is obtained by training based on the sample field image and the field weed identification result label corresponding to the sample field image, further the field weed identification model which is specially used for identifying the field image corresponding to the northern spring corn can be obtained by training, the sample field image used for training the model comprises a first dimension image, a second dimension image, a third dimension image, a fourth dimension image, a fifth dimension image and a sixth dimension image, the first dimension image comprises images corresponding to different growth stages of corn, the second dimension image comprises images corresponding to straw coverage and images corresponding to no straw coverage, the third dimension image comprises images corresponding to different cultivation ridge types, the fifth dimension image comprises images corresponding to different weeding treatment modes, the first dimension image, the fifth dimension image comprises images corresponding to the straw coverage images and the no straw coverage corresponding to the straw coverage image, the field image is fully considered, the characteristics of the field image is fully considered in the northern spring corn, the characteristics are fully identified, the characteristics of the field pattern is fully identified in consideration of the characteristics of the field pattern is fully identified in the northern spring corn, the characteristics are fully identified in the field pattern, the characteristics of the field pattern is fully identified in the northern spring corn, the field is fully identified, the characteristics of the field pattern is better than the field pattern is better compared with the characteristics of the field pattern is better than the field pattern compared with the field characteristics corresponding to the field pattern compared with the field pattern, and the field pattern is better than the characteristics corresponding to the field pattern compared with the field pattern is compared with the field pattern with the field images corresponding with the field pattern, the accuracy of variable plant protection operation can be improved to improve the plant protection effect, and then improve the output and the quality of northern spring corn.

Drawings

In order to more clearly illustrate the invention or the technical solutions of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are some embodiments of the invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic flow chart of a field weed identification method for northern spring corn provided by the invention;

fig. 2 is a schematic structural view of a field weed recognition device for northern spring corn provided by the invention;

fig. 3 is a schematic structural diagram of an electronic device provided by the present invention.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

In the planting process of northern spring corns, farmers commonly adopt chemical plant protection and weeding twice, and once is a closed weeding operation performed before emergence after sowing; and the other time is the stem leaf weeding operation performed when 3-5 leaves are spread after emergence. The field grass condition can be well controlled through twice chemical plant protection weeding, however, the plant protection operation mode has the phenomenon that pesticides are used excessively, so that the pollution to the field environment is caused, the plant protection cost is increased, and the planting cost is increased. Especially with the continuous increase of chemical herbicide prices, the planting cost is further increased.

At present, considering that the number of the post-emergence weeds is lower, the field non-differential spraying operation (namely the quantitative plant protection operation) is adopted, the pesticide waste is caused, the environmental pollution is increased, and the planting cost of farmers is increased at the same time. However, the current weed identification accuracy based on field images is not high, so that the accuracy of variable plant protection operation is reduced, the plant protection effect is finally affected, and the yield and quality of northern spring corns are reduced.

In view of the above problems, the present invention proposes the following embodiments. Fig. 1 is a schematic flow chart of a field weed identification method for northern spring corn, as shown in fig. 1, provided by the invention, comprising the following steps:

and 110, acquiring a field image of a northern spring corn planting area.

The invention is used for identifying field weeds in a field image of a planting area of northern spring corns, and the same effect cannot be achieved by identifying field weeds in other crops. In other words, the present invention is directed to improvements in northern spring corn.

Here, the northern spring corn planting area is a planting area where northern spring corn is planted, and the field image is an image corresponding to the planting area of northern spring corn, so that the image which is input to the field weed recognition model later is ensured to be an accurate image.

In a specific embodiment, the field image is acquired by an image acquisition device, which is a camera, for example.

In an embodiment, the field image is an image of a plant protection operation area, so that field weed identification is ensured to be performed on the area image of the plant protection operation area, whether weeds exist in the complete plant protection operation area or not is ensured to be judged, and the situation that some areas are not identified is avoided, so that the situation that some areas are not subjected to plant protection operation is avoided, and the accuracy of variable plant protection operation is improved.

In an embodiment, the field image is an image of a liquid medicine spraying area, so that field weed identification is performed on the area image of the liquid medicine spraying area, whether weeds exist in the liquid medicine spraying area or not is determined, the situation that some liquid medicine spraying areas are not identified is avoided, the situation that some areas are not subjected to plant protection operation is avoided, and accuracy of variable plant protection operation is improved. Further, the liquid medicine spraying area is determined by the operation width of the spraying device; of course, there may be a plurality of spraying devices, each of which corresponds to one of the field images, so as to control the opening and closing of each of the spraying devices based on the field weed recognition results corresponding to each of the field images, respectively.

And 120, inputting the field image into a field weed identification model to obtain a field weed identification result output by the field weed identification model.

Here, the field weed recognition result includes presence or absence of grass. Further, the field weed identification result includes corn with grass, corn without grass, no corn without grass or no corn with grass. It should be appreciated that the field weed identification result is an identification result of the field image corresponding area, and then it is preferable to determine whether the field image corresponding area is sprayed with the pesticide based on the field weed identification result.

In one embodiment, the field image is input to a field weed identification model corresponding to the northern spring corn planting area, and a field weed identification result output by the field weed identification model is obtained. Namely, different planting areas correspond to different field weed identification models so as to improve the field weed identification accuracy. It should be understood that the field weed identification models corresponding to the different planting areas require different training samples.

The field weed identification model is obtained through training based on a sample field image and a field weed identification result label corresponding to the sample field image.

The sample field image is a training sample of the field weed identification model, and the sample field image is also a field image of a northern spring corn planting area, so that the field weed identification model obtained through training is specially used for identifying the field image of the northern spring corn planting area, and further the field weed identification accuracy of the northern spring corn is improved.

In a specific embodiment, the sample field image is acquired by an image acquisition device, which is illustratively a camera.

Here, the field weed identification result label is a label obtained by labeling the sample field image. The field weed identification result label comprises weed or no weed. Further, the field weed identification result label comprises corn with grass, corn without grass or corn without grass. It is understood that the field weed identification result label is the actual weed condition of the area corresponding to the sample field image, so that the labeling accuracy of the field weed identification result label is improved, the training effect of the field weed identification model is further improved, and finally the field weed identification accuracy is improved.

The embodiment of the invention is characterized in that the sample field images corresponding to the northern spring corns are collected in 6 dimensions according to the characteristics of the northern spring corns. Further, according to the field cultivation characteristics of the northern spring corns, the method is divided into 6 dimensions to collect sample field images corresponding to the northern spring corns. Further, according to the characteristic of the northern spring corn under different planting conditions, the method is divided into 6 dimensions to collect sample field images corresponding to the northern spring corn.

In the embodiment of the invention, the images corresponding to the different growth stages of the northern spring corn are considered to have different degrees of influence on field weed identification, so that in order to ensure that the field weed identification model obtained by training has better identification rate on the images corresponding to the different growth stages of the corn, the images corresponding to the different growth stages of the corn are acquired in the training stage to carry out model training, thereby improving the model training effect and finally improving the field weed identification accuracy.

The first dimension image may include: at least one of an image corresponding to 2 expansion leaves, an image corresponding to 3 expansion leaves, an image corresponding to 4 expansion leaves, an image corresponding to 5 expansion leaves, and an image corresponding to 6 expansion leaves.

In one embodiment, the number of images of the first dimension is 500-1000. It is to be noted that, through verification, the number of images of the first dimension image is not less than 500, and the recognition rate of the field weed recognition model obtained through training can be larger than the preset recognition rate. The preset recognition rate is a training target of the model, for example, 85%. Further, the number of images of the first dimension image is 1000, so that the training effect of the field weed identification model is more optimal on the basis of ensuring the training efficiency.

In the embodiment of the invention, the straw of northern spring corns is considered to have a certain influence on field weed identification, for example, the straw is considered to be weed by mistake, so that in order to ensure that a field weed identification model obtained by training has better identification rate on whether straw is covered or not, an image corresponding to straw coverage and an image corresponding to no straw coverage are acquired in a training stage to carry out model training, thereby improving the model training effect and finally improving the field weed identification accuracy.

The image corresponding to the straw coverage indicates that the field soil area corresponding to the image comprises the straw coverage; the image corresponding to no straw coverage indicates that the field soil area corresponding to the image does not include straw coverage.

In one embodiment, the number of images of the second dimension is 500-1000. It should be noted that, through verification, the number of images of the second dimension image is not less than 500, and the recognition rate of the field weed recognition model obtained through training can be larger than the preset recognition rate. The preset recognition rate is a training target of the model, for example, 85%. Further, the number of images of the second dimension image is 1000, so that the training effect of the field weed identification model is more optimal on the basis of ensuring the training efficiency.

In the embodiment of the invention, the images corresponding to different corn cultivation ridge types of northern spring corn are considered to have different degrees of influence on field weed identification, so that in order to enable the field weed identification model obtained through training to have better identification rate on the images corresponding to different corn cultivation ridge types, the images corresponding to different corn cultivation ridge types are acquired in the training stage to carry out model training, the model training effect is improved, and finally the field weed identification accuracy is improved.

The third dimensional image may include, but is not limited to, at least one of: an image corresponding to ridge-free flat seeding (flat land cultivation), an image corresponding to ridging cultivation, and the like.

In consideration of the fact that images corresponding to different ridging cultivation modes of northern spring corns have different degrees of influence on field weed identification, in order to enable a field weed identification model obtained through training to have good identification rates on images corresponding to different ridging cultivation modes, images corresponding to different ridging cultivation modes are collected in a training stage to conduct model training, and therefore model training effects are improved, and finally field weed identification accuracy is improved.

The ridge-forming cultivation corresponding image may include, but is not limited to, at least one of: images corresponding to 2 seeding lines on a 110 cm wide ridge, images corresponding to 1 seeding line on a 65 cm wide ridge, images corresponding to 1 seeding line on a 75 cm wide ridge, and the like.

Wherein, 2 sowing lines on a 110 cm wide ridge means that two sowing lines exist on one 110 cm wide ridge, namely, two rows of corns are planted on one ridge.

In one embodiment, the number of images of the third dimension image is 500-1000. It should be noted that, through verification, the number of images of the third dimension image is not less than 500, and the recognition rate of the field weed recognition model obtained through training can be larger than the preset recognition rate. The preset recognition rate is a training target of the model, for example, 85%. Further, the number of images of the third dimension image is 1000, so that the training effect of the field weed identification model is more optimal on the basis of ensuring the training efficiency.

In the embodiment of the invention, the images corresponding to different soil types are taken into consideration, so that the images corresponding to the different soil types are planted in northern spring corn, and the effect of model training is improved, and finally the accuracy of field weed identification is improved.

Considering that the type of planting soil required for northern spring corn is mostly aeolian sandy soil, black lime soil, dark brown soil, white serosa soil, meadow soil, and the like, the fourth-dimensional image may include, but is not limited to, at least one of: an image corresponding to sand, an image corresponding to black lime, an image corresponding to dark brown soil, an image corresponding to white serosity, an image corresponding to meadow soil, and the like.

In one embodiment, the number of images of the fourth dimension is 500-1000. It should be noted that, through verification, the number of images of the fourth dimension image is not less than 500, and the recognition rate of the field weed recognition model obtained through training can be larger than the preset recognition rate. The preset recognition rate is a training target of the model, for example, 85%. Further, the number of images of the fourth dimension image is 1000, so that the training effect of the field weed identification model is more optimal on the basis of ensuring the training efficiency.

In the embodiment of the invention, the images corresponding to different weeding treatment modes are taken into consideration, so that different degrees of influence on field weed identification are realized, and therefore, in order to ensure that the field weed identification model obtained by training has better identification rate on the images corresponding to different weeding treatment modes, the images corresponding to different weeding treatment modes are acquired in the training stage to carry out model training, thereby improving the model training effect and finally improving the field weed identification accuracy.

Considering that the weeding treatment of northern spring corn is mostly a pre-seedling closed weeding treatment and a non-pre-seedling closed weeding treatment is only a post-seedling stem leaf weeding treatment, the fifth dimension image can comprise, but is not limited to, at least one of the following: an image corresponding to a pre-seedling closed weeding treatment mode only, an image corresponding to a post-seedling stem leaf weeding treatment mode without pre-seedling closed weeding treatment only, an image corresponding to a comprehensive treatment mode with pre-seedling closed weeding treatment and post-seedling stem leaf weeding treatment, and the like.

The comprehensive treatment modes of the pre-seedling closed weeding treatment and the post-seedling stem leaf weeding treatment are weeding modes adopted by the pre-seedling closed weeding treatment and the post-seedling stem leaf weeding treatment.

In one embodiment, the number of images of the fifth dimension image is 500-1000. It should be noted that, through verification, the number of images of the fifth dimension image is not less than 500, and the recognition rate of the field weed recognition model obtained through training can be larger than the preset recognition rate. The preset recognition rate is a training target of the model, for example, 85%. Further, the number of images of the fifth dimension image is 1000, so that the training effect of the field weed identification model is more optimal on the basis of ensuring the training efficiency.

In the embodiment of the invention, the images corresponding to different grass conditions are taken into consideration, and the images corresponding to different grass conditions are planted in northern spring corns, so that the field weed identification model obtained through training has good identification rate on the images corresponding to different grass conditions, and the images corresponding to different grass conditions are acquired in the training stage to perform model training, thereby improving the model training effect and finally improving the field weed identification accuracy.

In one embodiment, the number of images of the sixth dimension is 500-1000. It should be noted that, through verification, the number of images of the sixth dimension image is not less than 500, and the recognition rate of the field weed recognition model obtained through training can be larger than the preset recognition rate. The preset recognition rate is a training target of the model, for example, 85%. Further, the number of the sixth-dimension images is 1000, so that the training effect of the field weed identification model is more optimal on the basis of ensuring the training efficiency.

It should be noted that, through a great deal of creative labor and a great deal of experiments, a better training effect can be achieved by performing model training through the images with six dimensions. It can also be seen that the effect of the first dimension image on weed identification is greater than the effect of the second dimension image on weed identification; the second dimension image has a greater effect on weed identification than the third dimension image; the third dimension image has a greater effect on weed identification than the fourth dimension image; the fourth dimension image has a greater effect on weed recognition than the fifth dimension image; the effect of the fifth dimension image on weed recognition is greater than the effect of the sixth dimension image on weed recognition. Based on this, if the number of dimensional images having a low influence on weed recognition is small, the dimensional images having a high influence on weed recognition can be used for supplementing. From the above, it can be seen that the six-dimensional sample field image is obtained through a great deal of creative labor.

In one embodiment, considering that the first dimensional image has a greater effect on weed recognition than the second dimensional image, the second dimensional image has a greater effect on weed recognition than the third dimensional image, the fourth dimensional image has a greater effect on weed recognition than the fifth dimensional image, and the fifth dimensional image has a greater effect on weed recognition than the sixth dimensional image; based on the above, the number of images of the first dimension is larger than the number of images of the second dimension, the number of images of the second dimension is larger than the number of images of the third dimension, the number of images of the third dimension is larger than the number of images of the fourth dimension, the number of images of the fourth dimension is larger than the number of images of the fifth dimension, and the number of images of the fifth dimension is larger than the number of images of the sixth dimension; therefore, based on the method, sample field images are collected more reasonably, the training effect of a field weed identification model is improved, the identification accuracy of the field weed identification model is improved, namely the robustness of the field weed identification model is improved, and finally the field weed identification accuracy is improved, based on the method, the accuracy of variable plant protection operation can be improved, the plant protection effect is improved, and the yield and quality of northern spring corns are improved.

Further, after the step 120, on the basis of the field weed identification result, the opening and closing of the spraying device of the variable plant protection equipment is controlled so as to realize variable plant protection operation. In a specific embodiment, when the field weed identification result is that corn is grass or no corn is grass, controlling the spraying device to start so as to perform weeding operation; and under the condition that the field weeds are identified as corn without grass or corn without grass, controlling the spraying device to be closed so as to suspend weeding operation.

The method comprises the steps that an execution main body of the method is an edge calculation box of plant protection equipment, a spraying device comprises an electromagnetic valve and a spray head, and based on the edge calculation box outputs an open electric signal to the electromagnetic valve to enable the electromagnetic valve to be opened and further enable the spray head to perform spraying operation when corn grass exists or does not exist in a field weed identification result; under the condition that the field weeds are identified as corn grass or corn grass, the edge computing box outputs a closed electric signal to the electromagnetic valve, so that the electromagnetic valve is closed, and the spray head pauses the spraying operation.

It can be appreciated that by the mode, whether the field image has weeds or not is identified to determine whether pesticide is sprayed or not, so that variable plant protection operation (intelligent plant protection operation) is realized, namely differential spraying of post-seedling weeding operation of northern spring corn is realized. In the spraying process, the spray head can be opened in the grass-free area, and the spray head can be closed in the grass-free area, so that the pesticide spraying amount in the grass-free area is saved, the pesticide investment is reduced, the planting cost is reduced, and the environmental pollution is reduced.

It should be noted that, the training process of the field weed recognition model can be executed on the execution body of the method provided by the invention; the method can also be executed on a single training device, for example, the execution main body of the method provided by the invention is an edge calculation box, and the trained field weed identification model can be imported into the edge calculation box.

The field weed recognition method for northern spring corn provided by the embodiment of the invention inputs the field image of the northern spring corn planting area into the field weed recognition model to obtain the field weed recognition result output by the field weed recognition model, the field weed recognition model is obtained based on the sample field image and the field weed recognition result label corresponding to the sample field image, further the field weed recognition model which is specially used for recognizing the field image corresponding to the northern spring corn can be obtained through training, the sample field image used for training the model comprises a first dimension image, a second dimension image, a third dimension image, a fourth dimension image, a fifth dimension image and a sixth dimension image, the first dimension image comprises images corresponding to different growth stages of corn, the second dimension image comprises images corresponding to straw coverage and images corresponding to no straw coverage, the third dimension image comprises images corresponding to different corn cultivation ridge types, the fourth dimension image comprises images corresponding to different soil types, the fifth dimension image comprises images corresponding to different weeding treatment modes, the sixth dimension image comprises images corresponding to different grass conditions, so that the characteristics of the northern spring corn are fully considered, the field cultivation characteristics of the northern spring corn are fully considered, the characteristics of the northern spring corn under different planting conditions are fully considered, the sample field images corresponding to the northern spring corn are collected in 6 different dimensions, the training effect of a field weed identification model is improved, the identification accuracy of the field weed identification model is further improved, namely the robustness of the field weed identification model is improved, the field weed identification accuracy is finally improved, and the accuracy of variable plant protection operation can be improved based on the characteristics, so as to improve the plant protection effect and further improve the yield and quality of northern spring corns.

Based on any one of the above embodiments, in the method, the first dimension image includes an image corresponding to 2 expansion leaves, an image corresponding to 3 expansion leaves, an image corresponding to 4 expansion leaves, an image corresponding to 5 expansion leaves, and an image corresponding to 6 expansion leaves.

According to the field weed identification method for the northern spring corn, provided by the embodiment of the invention, the first dimension image for the training model comprises 2 images corresponding to the unfolding leaves, 3 images corresponding to the unfolding leaves, 4 images corresponding to the unfolding leaves, 5 images corresponding to the unfolding leaves and 6 images corresponding to the unfolding leaves, so that the characteristics of the northern spring corn are fully considered, different growth stages (sizes) of the northern spring corn are divided into 2 unfolding leaves, 3 unfolding leaves, 4 unfolding leaves, 5 unfolding leaves and 6 unfolding leaves, so that the images corresponding to different growth stages of the northern spring corn are collected in the training stage for model training, the training effect of the field weed identification model is improved, the identification accuracy of the field weed identification model is further improved, namely the robustness of the field weed identification model is improved, and finally the field weed identification accuracy is improved.

Based on any one of the above embodiments, in the method, the image corresponding to the covered straw includes an image corresponding to the irregularly stacked straw, and an image corresponding to the straw regularly stacked on both sides of the seeding row; the images corresponding to the irregularly stacked straws comprise a first irregular image and a second irregular image; the density of the straw in the acquisition area of the first irregular image is smaller than the preset density, and the density of the straw in the acquisition area of the second irregular image is larger than or equal to the preset density.

In one embodiment, the straw that is regularly stacked outside the straw on both sides of the sowing row may be determined as the irregularly stacked straw.

Here, the preset density may be set according to the actual situation, so as to divide the image corresponding to the irregularly stacked straw into the image corresponding to the irregularly stacked straw and the image corresponding to the irregularly stacked straw, which are sparsely stacked, based on the preset density.

According to the field weed identification method for northern spring corns, provided by the embodiment of the invention, the corresponding images for covering the straws of the training model comprise the images corresponding to the straws which are irregularly arranged, and the images corresponding to the straws which are regularly arranged at two sides of a sowing row, wherein the images corresponding to the straws which are irregularly arranged comprise the first irregular image and the second irregular image, the density of the straws in the acquisition area of the first irregular image is smaller than the preset density, the density of the straws in the acquisition area of the second irregular image is larger than or equal to the preset density, so that different arrangement modes of the straws of the northern spring corns are fully considered, the influence on field weed identification is caused to different degrees, the model training is carried out by acquiring the images corresponding to the different arrangement modes of the straws in the training stage, the training effect of the field weed identification model is further improved, the identification accuracy of the field weed identification model is further improved, namely the robustness of the field weed identification model is improved, and finally the field weed identification accuracy is improved; meanwhile, different stacking densities of the straws of the northern spring corns are fully considered, and influences on field weed identification are caused to different degrees, so that images corresponding to the different stacking densities of the straws are collected in a training stage to perform model training, the training effect of a field weed identification model is further improved, the identification accuracy of the field weed identification model is further improved, namely the robustness of the field weed identification model is improved, and finally the field weed identification accuracy is improved.

Based on any one of the above embodiments, in the method, the third dimensional image includes an image corresponding to ridgeless flat planting and an image corresponding to ridging cultivation.

According to the field weed identification method for the northern spring corn, provided by the embodiment of the invention, the third dimensional image for the training model comprises the image corresponding to the ridgeless flat planting and the image corresponding to the ridging cultivation, so that the characteristics of the northern spring corn are fully considered, the field cultivation characteristics of the northern spring corn are fully considered, the characteristics of the northern spring corn under different planting conditions are fully considered, the images corresponding to the ridging types of the northern spring corn are collected in the training stage to carry out model training, the training effect of the field weed identification model is improved, the identification accuracy of the field weed identification model is improved, namely the robustness of the field weed identification model is improved, and finally the field weed identification accuracy is improved.

Based on any embodiment, in the method, the image corresponding to the ridging cultivation comprises an image corresponding to 2 seeding lines on a 110 cm wide ridge, an image corresponding to 1 seeding line on a 65 cm wide ridge, and an image corresponding to 1 seeding line on a 75 cm wide ridge.

According to the field weed identification method for the northern spring corn, provided by the embodiment of the invention, the corresponding images for ridge forming cultivation of the training model comprise the images corresponding to 2 sowing lines on a 110 cm wide ridge, the images corresponding to 1 sowing line on a 65 cm wide ridge and the images corresponding to 1 sowing line on a 75 cm wide ridge, so that the characteristics of the northern spring corn are fully considered, the field cultivation characteristics of the northern spring corn are fully considered, the characteristics of the northern spring corn under different planting conditions are fully considered, so that the images corresponding to different ridge forming cultivation modes of the northern spring corn are collected in a training stage to perform model training, the training effect of the field weed identification model is improved, the identification accuracy of the field weed identification model is further improved, namely the robustness of the field weed identification model is improved, and finally the field weed identification accuracy is improved.

Based on any one of the above embodiments, in the method, the fourth dimension image includes an image corresponding to sand and wind soil, an image corresponding to black lime soil, an image corresponding to dark brown soil, an image corresponding to white serosity soil, and an image corresponding to meadow soil.

According to the field weed identification method for the northern spring corn, provided by the embodiment of the invention, the fourth dimension image for the training model comprises the image corresponding to the sand soil, the image corresponding to the black lime soil, the image corresponding to the dark brown soil, the image corresponding to the white slurry soil and the image corresponding to the meadow soil, so that the characteristics of the northern spring corn are fully considered, the field cultivation characteristics of the northern spring corn are fully considered, the characteristics of the northern spring corn under different planting conditions are fully considered, the images corresponding to different soil types of the northern spring corn are collected in the training stage to perform model training, the training effect of the field weed identification model is improved, the identification accuracy of the field weed identification model is further improved, namely the robustness of the field weed identification model is improved, and finally the field weed identification accuracy is improved.

Based on any one of the above embodiments, in the method, the fifth dimension image includes an image corresponding to only a pre-seedling closed weeding treatment mode, an image corresponding to only a post-seedling stem leaf weeding treatment mode without pre-seedling closed weeding treatment, and an image corresponding to a comprehensive treatment mode with pre-seedling closed weeding treatment and post-seedling stem leaf weeding treatment.

According to the field weed identification method for the northern spring corn, provided by the embodiment of the invention, the fifth dimension image for the training model comprises an image corresponding to a pre-seedling closed weeding treatment mode, an image corresponding to a post-seedling stem leaf weeding treatment mode without pre-seedling closed weeding treatment, and an image corresponding to a comprehensive treatment mode with pre-seedling closed weeding treatment and post-seedling stem leaf weeding treatment, so that the characteristics of the northern spring corn are fully considered, the field cultivation characteristics of the northern spring corn are fully considered, the characteristics of the northern spring corn under different planting conditions are fully considered, so that the images corresponding to different weeding treatment modes of the northern spring corn are collected in a training stage to perform model training, the training effect of the field weed identification model is improved, the identification accuracy of the field weed identification model is further improved, namely the robustness of the field weed identification model is improved, and finally the field weed identification accuracy is improved.

Based on any one of the above embodiments, in the method, the sixth-dimension image includes an image corresponding to a first grass condition degree, an image corresponding to a second grass condition degree, an image corresponding to a third grass condition degree, and an image corresponding to a fourth grass condition degree; the first grass plot degree is less than the second grass plot degree, the second grass plot degree is less than the third grass plot degree, and the third grass plot degree is less than the fourth grass plot degree.

Here, the first grass condition degree, the second grass condition degree, the third grass condition degree and the fourth grass condition degree are set according to the characteristics of the northern spring corn, the field cultivation characteristics of the northern spring corn and the characteristics of the northern spring corn under different planting conditions. Illustratively, the first grass condition level is mild, the second grass condition level is moderate, the third grass condition level is moderate, and the fourth grass condition level is severe.

According to the field weed identification method for the northern spring corn, the sixth dimension image for the training model comprises the image corresponding to the first grass condition degree, the image corresponding to the second grass condition degree, the image corresponding to the third grass condition degree and the image corresponding to the fourth grass condition degree, the first grass condition degree is smaller than the second grass condition degree, the second grass condition degree is smaller than the third grass condition degree, the third grass condition degree is smaller than the fourth grass condition degree, so that the characteristics of the northern spring corn are fully considered, the field cultivation characteristics of the northern spring corn are fully considered, the characteristics of the northern spring corn under different planting conditions are fully considered, the grass condition degree corresponding to the northern spring corn is divided into the first grass condition degree, the second grass condition degree, the third grass condition degree and the fourth grass condition degree, so that the model training is conducted by collecting the images corresponding to the northern spring corn at different grass condition degrees in a training stage, the field weed identification model is improved, the field weed identification model identification accuracy is improved, and finally the field weed identification accuracy is improved.

Based on any one of the above embodiments, in the method, the number of images of the first dimension image, the number of images of the second dimension image, the number of images of the third dimension image, the number of images of the fourth dimension image, the number of images of the fifth dimension image, and the number of images of the sixth dimension image are each 500 to 1000.

The field weed identification method for northern spring corns, provided by the embodiment of the invention, is used for training the number of images of a first dimension image, the number of images of a second dimension image, the number of images of a third dimension image, the number of images of a fourth dimension image, the number of images of a fifth dimension image and the number of images of a sixth dimension image of a model to be 500 to 1000, so that the training effect of the field weed identification model is better on the basis of ensuring the training efficiency, the identification accuracy of the field weed identification model is further improved, namely the robustness of the field weed identification model is improved, and finally the field weed identification accuracy is improved.

The field weed recognition device for northern spring corn provided by the invention is described below, and the field weed recognition device for northern spring corn described below and the field weed recognition method for northern spring corn described above can be correspondingly referred to each other.

Fig. 2 is a schematic structural diagram of a field weed identifying device for northern spring corn, as shown in fig. 2, the field weed identifying device for northern spring corn includes:

an acquisition module 210, configured to acquire a field image of a northern spring corn planting area;

the recognition module 220 is used for inputting the field image into a field weed recognition model to obtain a field weed recognition result output by the field weed recognition model;

The field weed recognition device for northern spring corn provided by the embodiment of the invention inputs the field image of the northern spring corn planting area into the field weed recognition model to obtain the field weed recognition result output by the field weed recognition model, the field weed recognition model is obtained based on the sample field image and the field weed recognition result label corresponding to the sample field image, further the field weed recognition model which is specially used for recognizing the field image corresponding to the northern spring corn can be obtained through training, the sample field image used for training the model comprises a first dimension image, a second dimension image, a third dimension image, a fourth dimension image, a fifth dimension image and a sixth dimension image, the first dimension image comprises images corresponding to different growth stages of corn, the second dimension image comprises images corresponding to straw coverage and images corresponding to no straw coverage, the third dimension image comprises images corresponding to different corn cultivation ridge types, the fourth dimension image comprises images corresponding to different soil types, the fifth dimension image comprises images corresponding to different weeding treatment modes, the sixth dimension image comprises images corresponding to different grass conditions, so that the characteristics of the northern spring corn are fully considered, the field cultivation characteristics of the northern spring corn are fully considered, the characteristics of the northern spring corn under different planting conditions are fully considered, the sample field images corresponding to the northern spring corn are collected in 6 different dimensions, the training effect of a field weed identification model is improved, the identification accuracy of the field weed identification model is further improved, namely the robustness of the field weed identification model is improved, the field weed identification accuracy is finally improved, and the accuracy of variable plant protection operation can be improved based on the characteristics, so as to improve the plant protection effect and further improve the yield and quality of northern spring corns.

Based on any of the above embodiments, the first dimension image includes an image corresponding to 2 expanded leaves, an image corresponding to 3 expanded leaves, an image corresponding to 4 expanded leaves, an image corresponding to 5 expanded leaves, and an image corresponding to 6 expanded leaves.

Based on any one of the above embodiments, the image corresponding to the covered straw includes an image corresponding to the irregularly stacked straw, and an image corresponding to the straw regularly stacked on both sides of the seeding row;

Based on any one of the above embodiments, the third dimensional image includes an image corresponding to ridgeless flat planting and an image corresponding to ridging cultivation.

Based on any of the above embodiments, the image corresponding to the ridging cultivation includes an image corresponding to 2 seeding lines on a 110 cm wide ridge, an image corresponding to 1 seeding line on a 65 cm wide ridge, and an image corresponding to 1 seeding line on a 75 cm wide ridge.

Based on any one of the above embodiments, the fourth dimension image includes an image corresponding to sand and wind soil, an image corresponding to black lime soil, an image corresponding to dark brown soil, an image corresponding to white serosity soil, and an image corresponding to meadow soil.

Based on any of the above embodiments, the fifth dimension image includes an image corresponding to only a pre-seedling closed weeding treatment mode, an image corresponding to only a post-seedling stem leaf weeding treatment mode without pre-seedling closed weeding treatment, and an image corresponding to a comprehensive treatment mode with pre-seedling closed weeding treatment and post-seedling stem leaf weeding treatment.

Based on any one of the above embodiments, the sixth-dimension image includes an image corresponding to a first grass condition degree, an image corresponding to a second grass condition degree, an image corresponding to a third grass condition degree, and an image corresponding to a fourth grass condition degree;

Based on any of the above embodiments, the number of images of the first dimension image, the number of images of the second dimension image, the number of images of the third dimension image, the number of images of the fourth dimension image, the number of images of the fifth dimension image, and the number of images of the sixth dimension image are each 500 to 1000.

Fig. 3 illustrates a physical schematic diagram of an electronic device, as shown in fig. 3, where the electronic device may include: processor 310, communication interface (Communications Interface) 320, memory 330 and communication bus 340, wherein processor 310, communication interface 320, memory 330 accomplish communication with each other through communication bus 340. Processor 310 may invoke logic instructions in memory 330 to perform a field weed identification method for northern spring corn, the method comprising: acquiring a field image of a northern spring corn planting area; inputting the field image into a field weed identification model to obtain a field weed identification result output by the field weed identification model; the field weed identification model is obtained by training based on a sample field image and a field weed identification result label corresponding to the sample field image; the sample field image comprises a first dimension image, a second dimension image, a third dimension image, a fourth dimension image, a fifth dimension image and a sixth dimension image; the first dimension image comprises images corresponding to different growth stages of corn; the second dimension image comprises an image corresponding to straw coverage and an image corresponding to no straw coverage; the third dimensional image comprises images corresponding to different corn cultivation ridge types; the fourth-dimension image comprises images corresponding to different soil types; the fifth dimension image comprises images corresponding to different weeding treatment modes; the sixth-dimension image comprises images corresponding to different grass conditions.

Further, the logic instructions in the memory 330 described above may be implemented in the form of software functional units and may be stored in a computer-readable storage medium when sold or used as a stand-alone product. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

In yet another aspect, the present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which when executed by a processor is implemented to perform the field weed identification method for northern spring corn provided by the above methods, the method comprising: acquiring a field image of a northern spring corn planting area; inputting the field image into a field weed identification model to obtain a field weed identification result output by the field weed identification model; the field weed identification model is obtained by training based on a sample field image and a field weed identification result label corresponding to the sample field image; the sample field image comprises a first dimension image, a second dimension image, a third dimension image, a fourth dimension image, a fifth dimension image and a sixth dimension image; the first dimension image comprises images corresponding to different growth stages of corn; the second dimension image comprises an image corresponding to straw coverage and an image corresponding to no straw coverage; the third dimensional image comprises images corresponding to different corn cultivation ridge types; the fourth-dimension image comprises images corresponding to different soil types; the fifth dimension image comprises images corresponding to different weeding treatment modes; the sixth-dimension image comprises images corresponding to different grass conditions.

The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.

From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on this understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective embodiments or some parts of the embodiments.

Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A field weed identification method for northern spring corn, comprising:

acquiring a field image of a northern spring corn planting area;

2. The field weed identification method for northern spring corn as claimed in claim 1, wherein the first dimension image comprises 2 images of developed leaves, 3 images of developed leaves, 4 images of developed leaves, 5 images of developed leaves and 6 images of developed leaves.

3. The method for identifying field weeds in northern spring corns according to claim 1, wherein the corresponding images of covered straws comprise irregularly stacked images corresponding to straws and regularly stacked images corresponding to straws on two sides of a sowing row;

4. The field weed identification method for northern spring corn as claimed in claim 1, wherein the third dimensional image comprises an image corresponding to no-ridge flat-seeding and an image corresponding to ridging cultivation.

5. The method for identifying weeds in a field for northern spring corn according to claim 4, wherein the image corresponding to the ridging cultivation comprises an image corresponding to 2 seeding lines on a 110 cm wide ridge, an image corresponding to 1 seeding line on a 65 cm wide ridge, and an image corresponding to 1 seeding line on a 75 cm wide ridge.

6. The field weed identification method for northern spring corn as claimed in claim 1, wherein the fourth dimension image comprises an image corresponding to sand, an image corresponding to black lime, an image corresponding to dark brown soil, an image corresponding to white serous soil and an image corresponding to meadow soil.

7. The method for identifying field weeds in northern spring corns according to claim 1, wherein the fifth dimension image comprises an image corresponding to a pre-seedling closed weeding treatment mode only, an image corresponding to a post-seedling stem leaf weeding treatment mode without pre-seedling closed weeding, and an image corresponding to a comprehensive treatment mode with pre-seedling closed weeding treatment and post-seedling stem leaf weeding treatment.

8. The field weed identification method for northern spring corn as claimed in claim 1, wherein the sixth-dimension image comprises an image corresponding to a first grass condition degree, an image corresponding to a second grass condition degree, an image corresponding to a third grass condition degree, and an image corresponding to a fourth grass condition degree;

9. The field weed identifying method for northern spring corn as claimed in any one of claims 1 to 8, wherein the number of images of the first dimension image, the number of images of the second dimension image, the number of images of the third dimension image, the number of images of the fourth dimension image, the number of images of the fifth dimension image and the number of images of the sixth dimension image are each 500 to 1000 sheets.

10. A field weed identification device for northern spring corn, comprising:

11. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor, when executing the program, implements the field weed identification method for northern spring corn as claimed in any one of claims 1 to 9.

12. A non-transitory computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor implements the field weed identification method for northern spring corn according to any one of claims 1 to 9.