CN111369494A - Method and device for detecting ear density of winter wheat - Google Patents

Abstract

Embodiments of the present invention provide a method and device for detecting ear density of winter wheat, the method comprising: acquiring a canopy image to be detected after the flowering period of winter wheat, and dividing the canopy image to be detected into several sub-images to be detected; Input the several sub-images to be detected into the convolutional neural network model, and output the sub-density map; determine the winter wheat ear density according to the sub-density map; wherein, the convolutional neural network model is based on the sample canopy image and the pre- The spike density map corresponding to the determined sample canopy image is obtained after training. The method and device for detecting ear density of winter wheat provided by the embodiments of the present invention utilize artificial intelligence technology for detection, with a high degree of automation, which can effectively reduce manual intervention in ear density estimation, reduce application cost and complexity, and effectively improve the accuracy of ear density estimation for winter wheat. and real-time.

Description

Method and device for detecting ear density of winter wheat

technical field

Embodiments of the present invention relate to the fields of systems engineering and information technology, and in particular, to a method and device for detecting ear density of winter wheat.

Background technique

The ear density of winter wheat is an important index for the estimation and phenotype analysis of winter wheat, and it has important practical significance. The traditional ear density calculation method mainly relies on manual counting. This method not only consumes a lot of manpower and material resources, but also has low efficiency and cannot meet the needs of large-area yield estimation and phenotypic analysis. The spike density estimation method based on computer vision has the advantages of low cost and easy use, and is currently the most important spike density estimation method. The computer vision-based ear density estimation method mainly includes the following steps: segmentation of wheat ears, morphological optimization of binary graphs, and calculation of the number of connected domains, and then the number of wheat ears in the image is obtained. Wheat ear image segmentation is the core content of computer vision-based ear density estimation methods, usually based on low-level features such as color and texture of wheat ears for pixel-level or superpixel-level wheat ear extraction.

Because the winter wheat canopy images collected in the field environment are easily disturbed by light conditions and complex backgrounds, the segmentation accuracy of wheat ear images based on the underlying features is easily disturbed by noise, and the generalization ability is low, so it is difficult to achieve good results in practical applications. Effect. In addition, the ear density in the winter wheat canopy images collected in the field environment is relatively high, so the mutual occlusion of the wheat ears is relatively serious, and the size and shape of the wheat ears change greatly, which brings challenges to the estimation of ear density. How to overcome the influence of light conditions and complex backgrounds, solve the problems that the current ear density estimation method is susceptible to noise interference, and the problems of wheat ear overlap and large changes in the appearance of wheat ears, and how to achieve accurate ear density estimation is an urgent problem to be solved.

SUMMARY OF THE INVENTION

Embodiments of the present invention provide a method and device for detecting ear density of winter wheat, which are used to solve the above-mentioned technical problems in the prior art.

In order to solve the above-mentioned technical problems, on the one hand, an embodiment of the present invention provides a method for detecting ear density of winter wheat, comprising:

Obtain the canopy image to be detected after the flowering period of winter wheat, and divide the canopy image to be detected into several sub-images to be detected;

Inputting the several sub-graphs to be detected into a convolutional neural network model, and outputting a sub-density graph;

Determine winter wheat ear density according to the sub-density map;

The convolutional neural network model is obtained after training based on the sample canopy image and the predetermined spike density map corresponding to the sample canopy image.

Further, determining the winter wheat ear density according to the sub-density map specifically includes:

Calculate the sub-density map position index according to the number of pooling layers in the convolutional neural network model;

According to the position index of the sub-density map, the output sub-density map is fused to obtain the ear density map;

The winter wheat ear density was determined from the ear density map.

Further, the training process of the convolutional neural network model includes the following steps:

Obtain the sample canopy image after the flowering period of winter wheat;

Use point markers to record the position information of wheat ears;

According to the position information of the wheat ears, the ear density map of the sample is generated;

A sliding window is used to divide the sample canopy image and its corresponding ear density map into several sample sub-maps and corresponding sub-density maps;

The convolutional neural network is trained by using the several sample sub-maps and the corresponding sub-density maps to obtain the convolutional neural network model.

Further, the position information of the wheat ear is the coordinates of the center point of the wheat ear in the image.

Further, according to the position information of the wheat ears, the ear density map of the sample is generated, which specifically includes:

According to the position information of wheat ears, the geometric adaptive method was used to generate the ear density map of the samples.

Further, the convolutional neural network is trained by using the several sample subgraphs and the corresponding sub-density graphs to obtain the convolutional neural network model, which specifically includes:

Using the several sample submaps and the corresponding subdensity maps as the input layer of the convolutional neural network model;

A feature extractor for building a convolutional neural network model;

And, sequentially connecting the input layer, the feature extractor, the overlay layer for fusing the output results of the feature extractor, and the output layer for fusing multiple channels in the overlay layer to complete the convolutional neural network model establishment.

Further, the feature extractor includes four processing modules in parallel, and each processing module includes three processing units connected in sequence.

Further, the first processing unit of each processing module includes a convolutional layer, a modified linear unit ReLU layer and a pooling layer that are connected in sequence; the second processing unit includes no less than 3 consecutively connected volumes. The accumulation layer and the modified linear unit ReLU layer, and only the first modified linear unit ReLU layer is connected to a pooling layer, and the third processing unit contains no less than 3 sequentially connected convolutional layers and modified linear units ReLU Floor;

Among them, the size of the convolution kernels in each convolutional layer in each processing unit remains unchanged, and the number of convolution kernels in each convolutional layer in the second processing unit is incremented by a multiplier of 2. The number of convolution kernels in each convolutional layer in the three processing units is successively decreased by a multiplier of 0.5, and the number of convolutional layers in the second processing unit is equal to the number of convolutional layers in the third processing unit, The size of the convolution kernels in the convolutional layers in the first processing unit is larger than the size of the convolutional kernels in each convolutional layer in the second processing unit, and the volume in the convolutional layers in the second processing unit The size of the accumulation kernel is larger than the size of the convolution kernels in each convolutional layer in the third processing unit.

Further, the size of the convolution kernel in the convolution layer in the first processing unit of each processing module decreases sequentially.

On the other hand, an embodiment of the present invention provides a device for detecting ear density of winter wheat, characterized in that it includes:

an acquisition module, configured to acquire the canopy image to be detected after the flowering period of winter wheat, and to divide the canopy image to be detected into several sub-images to be detected;

an output module for inputting the several sub-graphs to be detected into a convolutional neural network model, and outputting a sub-density graph;

a detection module for determining the winter wheat ear density according to the sub-density map;

The convolutional neural network model is obtained after training based on the sample canopy image and the predetermined spike density map corresponding to the sample canopy image.

The method and device for detecting ear density of winter wheat provided by the embodiments of the present invention utilize artificial intelligence technology for detection, with a high degree of automation, which can effectively reduce manual intervention in ear density estimation, reduce application cost and complexity, and effectively improve the accuracy of ear density estimation for winter wheat. and real-time.

Description of drawings

Fig. 1 is a schematic diagram of a method for detecting ear density of winter wheat provided by the embodiment of the present invention;

Fig. 2 is the winter wheat ear density detection logic flow chart provided by the embodiment of the present invention;

3 is a schematic diagram of a device for detecting ear density of winter wheat provided by an embodiment of the present invention;

FIG. 4 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention.

Detailed ways

In order to make the purposes, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments These are some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present invention.

FIG. 1 is a schematic diagram of a method for detecting ear density of winter wheat provided by an embodiment of the present invention. As shown in FIG. 1 , an embodiment of the present invention provides a method for detecting ear density of winter wheat, whose execution body is a device for detecting ear density of winter wheat. The method includes:

Step S101 , acquiring a canopy image to be detected after the flowering period of winter wheat, and dividing the canopy image to be detected into several sub-images to be detected.

Specifically, FIG. 2 is a logic flow chart for detecting ear density of winter wheat provided by an embodiment of the present invention. As shown in FIG. 2 , first, a canopy image to be detected after the flowering period of winter wheat is obtained, and the canopy image to be detected is divided into into several subgraphs to be detected.

For example, 2500×2500×3 pixels can be used to collect the canopy image to be detected, and the canopy image to be detected can be divided into several sub-images to be detected with a sliding window of 600×600 pixels and a step size of 300 pixels.

While dividing the to-be-detected canopy image into several to-be-detected sub-images, the location index of the to-be-detected sub-images is recorded.

The position index of the sub-image to be detected includes: the coordinates of the upper left corner pixel of the sub-image to be detected in the image to be detected, which can be extracted by using the first formula:

Among them, the first formula is:

Index _(i,j) =I _(i,j)

Index _(i,j) is the index of the jth sub-image of the i-th image to be detected, and I _(i,j) is the coordinates of the upper left pixel of the j-th sub-image of the i-th image to be detected in the image to be detected.

Step S102: Input the several sub-graphs to be detected into a convolutional neural network model, and output a sub-density map.

Specifically, after acquiring several sub-images to be detected, the several sub-images to be detected are input into the convolutional neural network model, and the sub-density map is output.

The convolutional neural network model is obtained after training based on a sample canopy image and a predetermined spike density map corresponding to the sample canopy image.

Step S103: Determine the winter wheat ear density according to the sub-density map.

Specifically, after the sub-density map is obtained, the winter wheat ear density is determined according to the sub-density map.

The method for detecting ear density of winter wheat provided by the embodiment of the present invention utilizes artificial intelligence technology for detection, has a high degree of automation, can effectively reduce manual intervention in ear density estimation, reduces application cost and complexity, and effectively improves the accuracy and accuracy of ear density estimation for winter wheat. real-time.

Based on any of the above-mentioned embodiments, further, determining the winter wheat ear density according to the sub-density map specifically includes:

Calculate the sub-density map position index according to the number of pooling layers in the convolutional neural network model;

According to the position index of the sub-density map, the output sub-density map is fused to obtain the ear density map;

The winter wheat ear density was determined from the ear density map.

Specifically, in the embodiment of the present invention, determining the ear density of winter wheat according to the sub-density map specifically includes the following steps:

First, the sub-density map position index is calculated according to the number of pooling layers in the convolutional neural network model.

Wherein, calculating the position index of the sub-density map to be detected, firstly, the second formula is used to extract the correction factor, and then the third formula is used to extract the position index of the sub-density map to be detected;

Wherein, the second formula is:

f is a correction factor, and N is the number of pooling layers in the convolutional neural network.

The third formula is:

Index_m _(i,j) =f*Index _(i,j)

Index _(i,j) is the index of the jth sub-map of the i-th image to be detected, and Index_m _(i,j) is the index of the j-th sub-density map of the i-th image to be detected.

Then, the output sub-density maps are fused according to the position index of the sub-density map to obtain the ear density map.

Finally, winter wheat ear density was determined from the ear density map.

The method for detecting ear density of winter wheat provided by the embodiment of the present invention utilizes artificial intelligence technology for detection, has a high degree of automation, can effectively reduce manual intervention in ear density estimation, reduces application cost and complexity, and effectively improves the accuracy and accuracy of ear density estimation for winter wheat. real-time.

Based on any of the above embodiments, further, the training process of the convolutional neural network model includes the following steps:

Obtain the sample canopy image after the flowering period of winter wheat;

Use point markers to record the position information of wheat ears;

According to the position information of the wheat ears, the ear density map of the sample is generated;

A sliding window is used to divide the sample canopy image and its corresponding ear density map into several sample sub-maps and corresponding sub-density maps;

The convolutional neural network is trained by using the several sample sub-maps and the corresponding sub-density maps to obtain the convolutional neural network model.

Specifically, the canopy images of winter wheat after the flowering stage collected in the field environment were firstly obtained as samples.

After the sample canopy image is acquired, preprocessing can also be performed, and the preprocessing is to adjust the size of the sample canopy image, for example, to 2500×2500×3 pixels.

Then, the position information of the wheat ears is recorded using dot marks.

In a specific application, preferably, the position information of the wheat ear includes: the coordinates of the center point of the wheat ear in the image.

Then, according to the position information of the wheat ears, the ear density map of the sample is generated.

Wherein, the ear density map of the sample is generated by a geometric adaptive method.

Then, a sliding window is used to divide the sample canopy image and its corresponding ear density map into several sample sub-maps and corresponding sub-density maps.

A sliding window sample image with a size of 600×600 pixels and a step size of 300 pixels and its corresponding spike density map can be divided into sub-images and corresponding sub-density maps.

Finally, the convolutional neural network is trained by using the several sample subgraphs and the corresponding sub-density graphs to obtain the convolutional neural network model.

In a specific application, preferably, the convolutional neural network structure includes 1 input layer, 1 feature extractor, 1 stacking layer and 1 output layer. The feature extractor includes four parallel processing modules, and each processing module includes three processing units connected in sequence.

In a specific application, preferably, the first processing unit of the first processing module includes a convolution layer, a modified linear unit ReLU layer and a pooling layer. Among them, 8 11×11 convolution kernels are used in the convolutional layer, and 2×2 maximum pooling function is used in the pooling layer; the second processing unit includes 3 convolutional layers and 3 modified linear unit ReLU layers and 1 pooling layer. Among them, 9 × 9 convolution kernels are used in the convolution layer, the number is 16, 32 and 64 respectively, the pooling layer is located after the first modified linear unit ReLU layer, and the maximum pooling function of 2 × 2 is used; The 3 processing units include 3 convolutional layers. Among them, 1×1 convolution kernels are used in the convolution layer, and the number is 32, 16 and 8 respectively.

In a specific application, preferably, the first processing unit of the second processing module includes a convolution layer, a modified linear unit ReLU layer and a pooling layer. Among them, 16 convolution kernels of 9 × 9 are used in the convolution layer, and the maximum pooling function of 2 × 2 is used in the pooling layer; the second processing unit includes 3 convolution layers and 3 modified linear units ReLU layers and 1 pooling layer. Among them, 7 × 7 convolution kernels are used in the convolution layer, the number is 32, 64 and 128 respectively, the pooling layer is located after the first modified linear unit ReLU layer, and the maximum pooling function of 2 × 2 is used; The 3 processing units include 3 convolutional layers. Among them, 1×1 convolution kernels are used in the convolution layer, and the number is 64, 32 and 16 respectively.

In a specific application, preferably, the first processing unit of the third processing module includes a convolution layer, a modified linear unit ReLU layer and a pooling layer. Among them, 32 7×7 convolution kernels are used in the convolutional layer, and 2×2 maximum pooling function is used in the pooling layer; the second processing unit includes 3 convolutional layers and 3 modified linear unit ReLU layers and 1 pooling layer. Among them, 5 × 5 convolution kernels are used in the convolution layer, the number is 64, 128 and 256 respectively, the pooling layer is located after the first modified linear unit ReLU layer, and the maximum pooling function of 2 × 2 is used; The 3 processing units include 3 convolutional layers. Among them, 1×1 convolution kernels are used in the convolution layer, and the number is 128, 64 and 32 respectively.

In a specific application, preferably, the first processing unit of the fourth processing module includes a convolution layer, a modified linear unit ReLU layer and a pooling layer. Among them, 64 5×5 convolution kernels are used in the convolution layer, and 2×2 maximum pooling function is used in the pooling layer; the second processing unit includes 3 convolution layers and 3 modified linear units ReLU layers and 1 pooling layer. Among them, 3 × 3 convolution kernels are used in the convolution layer, the number is 128, 256 and 512 respectively, the pooling layer is located after the first modified linear unit ReLU layer, and the maximum pooling function of 2 × 2 is used; The 3 processing units include 3 convolutional layers. Among them, 1×1 convolution kernels are used in the convolution layer, and the number is 256, 128 and 64 respectively.

In a specific application, preferably, the overlay layer superimposes the feature maps of the results of the four processing modules; the output layer includes a convolution layer, and a 1×1 convolution kernel is used to fuse the feature maps output by the overlay layer, thereby Output spike density map.

In the method for detecting ear density of winter wheat provided by the embodiment of the present invention, the canopy images collected in the field environment after the flowering period of winter wheat are preprocessed, the preprocessed images are divided into sample sets and to-be-detected sets, and point marks are used to record The position information of wheat ears; according to the position information of wheat ears, the ear density map of the sample set image is generated; using a sliding window, the winter wheat canopy image and its corresponding ear density map are divided into sub-maps and corresponding sub-density maps, and the The detection set image is divided into sub-images, and the position index of the sub-images of the set to be detected is recorded; the sample set sub-image and the corresponding sub-density map data are used as the input of the model, and the convolutional neural network model for estimating the ear density is constructed. The adaptive moment estimation algorithm is used for model training and verification; the sub-graph of the set to be tested is used for testing, and the sub-density map of the set to be detected is obtained; according to the number of pooling layers in the convolutional neural network model, the position of the sub-density map of the to-be-detected set is calculated index; according to the position index of the sub-density map of the set to be detected, the generated sub-density map of the set to be detected is fused to obtain the image ear density map of the to-be-detected set, and the corresponding ear density is further obtained. It can effectively reduce the manual intervention of ear density estimation, reduce the application cost and complexity, and effectively improve the accuracy and real-time performance of winter wheat ear density estimation. On the basis of the combination of the present invention with image processing, deep learning and other counting, it will make a great contribution to the estimation of the ear density of winter wheat.

Based on any of the above embodiments, FIG. 3 is a schematic diagram of an apparatus for detecting ear density of winter wheat provided by an embodiment of the present invention. As shown in FIG. 3 , an embodiment of the present invention provides an apparatus for detecting ear density of winter wheat, including an acquisition module 301 and an output module 302 and detection module 303, wherein:

The acquisition module 301 is used to acquire the to-be-detected canopy image after the flowering period of winter wheat, and to divide the to-be-detected canopy image into several sub-images to be detected; the output module 302 is used to input the several sub-images to be detected to the convolutional neural network model, and output the sub-density map; the detection module 303 is used to determine the winter wheat ear density according to the sub-density map; wherein, the convolutional neural network model is based on the sample canopy image and the predetermined sample canopy The spike density map corresponding to the image is obtained after training.

An embodiment of the present invention provides an apparatus for detecting ear density of winter wheat, which is used to execute the method described in any of the foregoing embodiments. The corresponding embodiments are the same, and are not repeated here.

The winter wheat ear density detection device provided by the embodiment of the present invention uses artificial intelligence technology for detection, has a high degree of automation, can effectively reduce manual intervention in ear density estimation, reduce application costs and complexity, and effectively improve the accuracy and efficiency of winter wheat ear density estimation. real-time.

FIG. 4 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. As shown in FIG. 4 , the electronic device includes: a processor (processor) 401, a communications interface (Communications Interface) 402, a memory (memory) 403, and a communication bus 404 , wherein, the processor 401 , the communication interface 402 , and the memory 403 complete the communication with each other through the communication bus 404 . The processor 401 and the memory 402 communicate with each other through the bus 403 . The processor 401 may invoke logic instructions in the memory 403 to perform the following methods:

Obtain the canopy image to be detected after the flowering period of winter wheat, and divide the canopy image to be detected into several sub-images to be detected;

Inputting the several sub-graphs to be detected into a convolutional neural network model, and outputting a sub-density graph;

Determine winter wheat ear density according to the sub-density map;

The convolutional neural network model is obtained after training based on the sample canopy image and the predetermined spike density map corresponding to the sample canopy image.

In addition, the above-mentioned logic instructions in the memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention can be embodied in the form of a software product in essence, or the part that contributes to the prior art or the part of the technical solution. The computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, Read-Only Memory (ROM, Read-Only Memory), Random Access Memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program codes .

Further, an embodiment of the present invention provides a computer program product, the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, and when the program instructions are executed by a computer When executed, the computer can execute the steps in the above method embodiments, for example, including:

Obtain the canopy image to be detected after the flowering period of winter wheat, and divide the canopy image to be detected into several sub-images to be detected;

Inputting the several sub-graphs to be detected into a convolutional neural network model, and outputting a sub-density graph;

Determine winter wheat ear density according to the sub-density map;

The convolutional neural network model is obtained after training based on the sample canopy image and the predetermined spike density map corresponding to the sample canopy image.

Further, an embodiment of the present invention provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiments are implemented, for example, including:

Obtain the canopy image to be detected after the flowering period of winter wheat, and divide the canopy image to be detected into several sub-images to be detected;

Inputting the several sub-graphs to be detected into a convolutional neural network model, and outputting a sub-density graph;

Determine winter wheat ear density according to the sub-density map;

The convolutional neural network model is obtained after training based on the sample canopy image and the predetermined spike density map corresponding to the sample canopy image.

The device embodiments described above are only illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in One place, or it can be distributed over multiple network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution in this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.

From the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and certainly can also be implemented by hardware. Based on this understanding, the above-mentioned technical solutions can be embodied in the form of software products in essence or the parts that make contributions to the prior art, and the computer software products can be stored in computer-readable storage media, such as ROM/RAM, magnetic A disc, an optical disc, etc., includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform the methods described in various embodiments or some parts of the embodiments.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that it can still be The technical solutions described in the foregoing embodiments are modified, or some technical features thereof are equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (10)

1. a winter wheat ear density detection method, is characterized in that, comprises: Obtain the canopy image to be detected after the flowering period of winter wheat, and divide the canopy image to be detected into several sub-images to be detected; Inputting the several sub-graphs to be detected into a convolutional neural network model, and outputting a sub-density graph; Determine winter wheat ear density according to the sub-density map; The convolutional neural network model is obtained after training based on the sample canopy image and the predetermined spike density map corresponding to the sample canopy image. 2. winter wheat ear density detection method according to claim 1, is characterized in that, described according to described sub-density map to determine winter wheat ear density, specifically comprises: Calculate the sub-density map position index according to the number of pooling layers in the convolutional neural network model; According to the position index of the sub-density map, the output sub-density map is fused to obtain the ear density map; The winter wheat ear density was determined from the ear density map. 3. winter wheat ear density detection method according to claim 1, is characterized in that, the training process of described convolutional neural network model comprises the steps: Obtain the sample canopy image after the flowering period of winter wheat; Use point markers to record the position information of wheat ears; According to the position information of the wheat ears, the ear density map of the sample is generated; A sliding window is used to divide the sample canopy image and its corresponding ear density map into several sample sub-maps and corresponding sub-density maps; The convolutional neural network is trained by using the several sample sub-maps and the corresponding sub-density maps to obtain the convolutional neural network model. 4 . The method for detecting ear density of winter wheat according to claim 3 , wherein the position information of the wheat ear is the coordinates of the center point of the wheat ear in the image. 5 . 5. winter wheat ear density detection method according to claim 3, is characterized in that, described according to the position information of wheat ear, generate the ear density map of sample, specifically comprises: According to the position information of wheat ears, the geometric adaptive method was used to generate the ear density map of the samples. 6. winter wheat ear density detection method according to claim 3, is characterized in that, described utilizes described several sample sub-graphs and corresponding sub-density graphs to carry out training to convolutional neural network, obtain described convolutional neural network models, including: Using the several sample submaps and the corresponding subdensity maps as the input layer of the convolutional neural network model; A feature extractor for building a convolutional neural network model; And, sequentially connecting the input layer, the feature extractor, the overlay layer for fusing the output results of the feature extractor, and the output layer for fusing multiple channels in the overlay layer to complete the convolutional neural network model establishment. 7 . The method for detecting ear density of winter wheat according to claim 6 , wherein the feature extractor comprises four parallel processing modules, and each processing module comprises three processing units connected in sequence. 8 . 8. winter wheat ear density detection method according to claim 7 is characterized in that, in the first processing unit of each processing module, all comprise successively connected convolution layer, modified linear unit ReLU layer and pooling layer; Both processing units contain no less than 3 convolutional layers and modified linear unit ReLU layers connected in sequence, and only the first modified linear unit ReLU layer is connected to a pooling layer, and the third processing unit contains No less than 3 consecutively connected convolutional layers and modified linear unit ReLU layers; Among them, the size of the convolution kernels in each convolutional layer in each processing unit remains unchanged, and the number of convolution kernels in each convolutional layer in the second processing unit is incremented by a multiplier of 2. The number of convolution kernels in each convolutional layer in the three processing units is successively decreased by a multiplier of 0.5, and the number of convolutional layers in the second processing unit is equal to the number of convolutional layers in the third processing unit, The size of the convolution kernels in the convolutional layers in the first processing unit is larger than the size of the convolutional kernels in each convolutional layer in the second processing unit, and the volume in the convolutional layers in the second processing unit The size of the accumulation kernel is larger than the size of the convolution kernels in each convolutional layer in the third processing unit. 9 . The method for detecting ear density of winter wheat according to claim 8 , wherein the size of the convolution kernel in the convolution layer in the first processing unit of each processing module decreases sequentially. 10 . 10. a winter wheat ear density detection device, is characterized in that, comprises: an acquisition module, configured to acquire the canopy image to be detected after the flowering period of winter wheat, and to divide the canopy image to be detected into several sub-images to be detected; an output module for inputting the several sub-graphs to be detected into a convolutional neural network model, and outputting a sub-density graph; a detection module for determining the winter wheat ear density according to the sub-density map; The convolutional neural network model is obtained after training based on the sample canopy image and the predetermined spike density map corresponding to the sample canopy image.

Images