CN113469064B - Identification method and system for corn leaf disease image in complex environment - Google Patents
Identification method and system for corn leaf disease image in complex environment Download PDFInfo
- Publication number
- CN113469064B CN113469064B CN202110757395.XA CN202110757395A CN113469064B CN 113469064 B CN113469064 B CN 113469064B CN 202110757395 A CN202110757395 A CN 202110757395A CN 113469064 B CN113469064 B CN 113469064B
- Authority
- CN
- China
- Prior art keywords
- convolution kernel
- feature map
- convolution
- feature
- size
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 240000008042 Zea mays Species 0.000 title claims abstract description 66
- 235000002017 Zea mays subsp mays Nutrition 0.000 title claims abstract description 66
- 201000010099 disease Diseases 0.000 title claims abstract description 64
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 title claims abstract description 64
- 235000005824 Zea mays ssp. parviglumis Nutrition 0.000 title claims abstract description 59
- 235000005822 corn Nutrition 0.000 title claims abstract description 59
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000011176 pooling Methods 0.000 claims abstract description 27
- 238000010586 diagram Methods 0.000 claims abstract description 7
- 230000004913 activation Effects 0.000 claims description 60
- 238000012549 training Methods 0.000 claims description 36
- 230000010339 dilation Effects 0.000 claims description 24
- 238000012360 testing method Methods 0.000 claims description 16
- 235000016383 Zea mays subsp huehuetenangensis Nutrition 0.000 claims description 7
- 235000009973 maize Nutrition 0.000 claims description 7
- 241000700605 Viruses Species 0.000 claims description 4
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 abstract description 14
- 230000004927 fusion Effects 0.000 abstract description 11
- 230000007547 defect Effects 0.000 abstract description 4
- 230000006870 function Effects 0.000 description 94
- 230000008569 process Effects 0.000 description 5
- 238000002474 experimental method Methods 0.000 description 4
- 230000007246 mechanism Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 241000196324 Embryophyta Species 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000008094 contradictory effect Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000013154 diagnostic monitoring Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the invention provides a method and a system for identifying corn leaf disease images in a complex environment, and belongs to the technical field of image identification. The identification method comprises the following steps: acquiring a corn leaf disease image to be identified; performing a first convolution operation on the image to obtain a first feature map; performing a first pooling operation on the first feature map according to formula (1) to obtain a second feature map; performing a second convolution operation on the second feature map to obtain a third feature map; performing a second pooling operation on the third feature map according to formula (2) to obtain a fourth feature map; and inputting the fourth characteristic diagram into a classification layer to determine the corn leaf disease type corresponding to the image. The method and the system realize accurate and efficient identification of the corn leaf disease image by adopting the lightweight neural network by constructing the lightweight neural network and combining with the fusion principle of the CASF characteristics, and overcome the technical defect of application scene limitation of the identification method caused by adopting the weight neural network in the prior art.
Description
Technical Field
The invention relates to the technical field of image recognition, in particular to a method and a system for recognizing corn leaf disease images in a complex environment.
Background
Corn diseases are mostly reflected on leaves, and corn leaf diseases indirectly affect the yield and quality of corn, so that a skilled attendant can make corresponding disease judgment by observing corn leaves in the traditional sense, but the work is time-consuming and subject to subjective influence. The method is characterized in that the maize growth vigor and maize leaf disease monitoring are managed in a brand-new informatization mode in a large scale, and disease judgment is timely and effectively made, so that the yield and the quantity are ensured, and the method is a necessary choice for future agricultural development. Computer vision and image processing techniques have found widespread use in diagnostic monitoring applications in the agricultural field, such as plant species classification, leaf disease identification, and plant disease severity estimation.
Effective and accurate identification of corn leaf diseases with multiple scales in complex environments (including different scenes) is often limited, and the identification accuracy is not high. The corn leaf disease image obtained in multiple scenes is usually provided with complex backgrounds, and the backgrounds are easy to interfere with disease characteristics, so that the difficulty of accurately identifying corn leaf diseases is increased, and compared weight type networks are mostly adopted for identifying corn leaf diseases in the complex environments in the aspect of deep learning. The recognition accuracy of the weight type network is relatively objective, but the recognition speed is low, so that the application scene of corn disease image recognition is restricted.
Disclosure of Invention
The invention aims to provide a method and a system for identifying corn leaf disease images in a complex environment, which expand and improve the application scene of the method for identifying corn leaf disease images.
In order to achieve the above object, an embodiment of the present invention provides a method for identifying a corn leaf disease image in a complex environment, the method comprising:
acquiring a corn leaf disease image to be identified;
performing a first convolution operation on the image to obtain a first feature map;
performing a first pooling operation on the first feature map according to equation (1) to obtain a second feature map,
f pool ＝
Max(x m,n ,x m,n+1 ,x m,n+2 ,x m+1,n ,x m+1,n+1 ,x m+1,n+2 ,x m+2,n ,x m+2,n+1 ,x m+2,n+2 )，(1)
wherein f pool X is a pixel value in the feature map, and m and n are position information of pixels;
performing a second convolution operation on the second feature map to obtain a third feature map, wherein the second convolution operation includes:
inputting the second feature map into a first convolution kernel to obtain a first CASF feature;
inputting the second feature map and the first CASF feature into a second convolution kernel to obtain a second CASF feature;
inputting the first and second CASF features into a third convolution kernel to obtain a third CASF feature;
inputting the third CASF feature into a fourth convolution kernel to obtain the third feature map; performing a second pooling operation on the third feature map according to formula (2) to obtain a fourth feature map;
wherein y is c For the c-th channel x c The output of (i, j), W being the width value of the input feature and H being the height value of the input feature;
and inputting the fourth characteristic diagram into a classification layer to determine the corn leaf disease type corresponding to the image.
Optionally, the first convolution kernel, the second convolution kernel, and the third convolution kernel each include three convolution kernels in series in sequence, wherein:
the first convolution kernel of the first convolution kernel has a size of 1*1, a dilation rate of 1, an activation function of a ReLU function, a second convolution kernel has a size of 3*3, a dilation rate of 1, an activation function of a ReLU function, and a third convolution kernel has a size of 1*1, a dilation rate of 1, and an activation function of a ReLU function;
the first convolution kernel of the second convolution kernel has a size of 1*1, an expansion rate of 2, an activation function of a ReLU function, the second convolution kernel has a size of 3*3, an expansion rate of 2, an activation function of a ReLU function, and the third convolution kernel has a size of 1*1, an expansion rate of 2, and an activation function of a ReLU function;
the first convolution kernel of the third convolution kernel has a size of 1*1, an expansion rate of 3, an activation function of a ReLU function, a second convolution kernel has a size of 3*3, an expansion rate of 3, an activation function of a ReLU function, a third convolution kernel has a size of 1*1, an expansion rate of 3, and an activation function of a ReLU function;
the fourth convolution kernel includes a convolution kernel of size 3*3, step size 2, and activation function ReLU.
Optionally, the performing a second convolution operation on the second feature map to obtain a third feature map includes:
and executing the second convolution operation for three times, wherein after any second convolution operation, the number of channels of the feature map after operation is twice the number of channels of the feature map before operation, and the scale of the feature map after operation is half of the scale of the feature map before operation.
Optionally, inputting the fourth feature map into a classification layer to determine a corn leaf disease type corresponding to the image includes:
inputting the fourth feature map into a classification layer to obtain the probability of each corn leaf disease type corresponding to the image;
and selecting the maize virus type with the highest probability as a final recognition result.
On the other hand, the invention also provides a system for identifying corn leaf disease images in a complex environment, which comprises:
the input layer is used for acquiring a corn leaf disease image to be identified;
the first convolution layer is used for executing first convolution operation on the image to obtain a first feature map;
a first pooling layer for performing a first pooling operation on the first feature map according to formula (1) to obtain a second feature map,
f pool ＝
Max(x m,n ,x m,n+1 ,x m,n+2 ,x m+1,n ,x m+1,n+1 ,x m+1,n+2 ,x m+2,n ,x m+2,n+1 ,x m+2,n+2 )，(1)
wherein f pool X is a pixel value in the feature map, and m and n are position information of pixels;
a second convolution layer, configured to perform a second convolution operation on the second feature map to obtain a third feature map, where the second convolution layer includes:
a first convolution kernel for generating a first CASF feature from the second feature map;
a second convolution kernel, configured to obtain a second CASF feature according to the second feature map and the first CASF feature;
a third convolution kernel configured to obtain a third CASF feature according to the first CASF feature and the second CASF feature;
a fourth convolution kernel, configured to obtain the third feature map according to the third CASF feature;
a second pooling layer for performing a second pooling operation on the third feature map according to formula (2) to obtain a fourth feature map;
wherein y is c For the c-th channel x c The output of (i, j), W being the width value of the input feature and H being the height value of the input feature;
and the classifying layer is used for inputting the fourth characteristic diagram into the classifying layer to determine the corn leaf disease type corresponding to the image.
Optionally, the first convolution kernel, the second convolution kernel, and the third convolution kernel each include three convolution kernels in series in sequence, wherein:
the first convolution kernel of the first convolution kernel has a size of 1*1, a dilation rate of 1, an activation function of a ReLU function, a second convolution kernel has a size of 3*3, a dilation rate of 1, an activation function of a ReLU function, and a third convolution kernel has a size of 1*1, a dilation rate of 1, and an activation function of a ReLU function;
the first convolution kernel of the second convolution kernel has a size of 1*1, an expansion rate of 2, an activation function of a ReLU function, the second convolution kernel has a size of 3*3, an expansion rate of 2, an activation function of a ReLU function, and the third convolution kernel has a size of 1*1, an expansion rate of 2, and an activation function of a ReLU function;
the first convolution kernel of the third convolution kernel has a size of 1*1, an expansion rate of 3, an activation function of a ReLU function, a second convolution kernel has a size of 3*3, an expansion rate of 3, an activation function of a ReLU function, a third convolution kernel has a size of 1*1, an expansion rate of 3, and an activation function of a ReLU function;
the fourth convolution kernel includes a convolution kernel of size 3*3, step size 2, and activation function ReLU.
Optionally, the second convolution layer is three layers, each layer includes two first convolution kernels, one second convolution kernel and one fourth convolution kernel, wherein the number of channels of the output feature map of each layer of the second convolution layer is twice the number of channels of the input feature map, and the scale of the output feature map is half of the scale of the input feature map.
Optionally, the classification layer is configured to:
generating probability of each corn leaf disease type corresponding to the image according to the fourth feature map;
and selecting the maize virus type with the highest probability as a final recognition result.
In yet another aspect, the present invention further provides a training method for training an identification system as described in any one of the above, including:
presetting a training set and a testing set;
training the recognition system by adopting the training set;
testing the identification system using the test set;
calculating errors of the classification result output by the identification system and the standard result corresponding to the test image of the test set;
judging whether the error is smaller than a preset error threshold value or not;
in the case where the error is determined to be greater than or equal to the error threshold, updating the parameters of the identification system using equation (3),
l asce ＝p y 2 log(p y )l cce +(1-p y )l rce ， (3)
wherein l asce For adaptive cross entropy loss function, p y To output probability, l cce For the forward cross entropy loss, l rce Is the inverse cross entropy loss;
training the recognition system by adopting the training set again, and executing corresponding steps of the training method until the error is judged to be smaller than the error threshold;
and outputting the recognition system after training is completed under the condition that the error is judged to be smaller than the error threshold value.
In yet another aspect, the invention also provides a computer readable storage medium, characterized in that the computer readable storage medium stores instructions for reading by a machine to cause the machine to perform a method as described in any one of the above.
By the technical scheme, the identification method and the identification system for the corn leaf disease image in the complex environment realize accurate and efficient identification of the corn leaf disease image by adopting the light-weight neural network by constructing the light-weight neural network and combining with the fusion principle of the CASF characteristics, and overcome the technical defect of limitation of the application scene of the identification method caused by adopting the light-weight neural network in the prior art.
Additional features and advantages of embodiments of the invention will be set forth in the detailed description which follows.
Drawings
The accompanying drawings are included to provide a further understanding of embodiments of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain, without limitation, the embodiments of the invention. In the drawings:
FIG. 1 is a flow chart of a method of identifying corn leaf disease images in a complex environment according to one embodiment of the invention;
FIG. 2 is a schematic diagram of each second convolution layer in accordance with one embodiment of the present disclosure;
FIG. 3 is a schematic diagram of a maize leaf disease image recognition system in a complex environment according to one embodiment of the present invention;
fig. 4 is a flow chart of a training method according to one embodiment of the invention.
Detailed Description
The following describes specific embodiments of the present invention in detail with reference to the drawings. It should be understood that the detailed description and specific examples, while indicating and illustrating the invention, are not intended to limit the invention.
In the embodiments of the present invention, unless otherwise indicated, terms of orientation such as "upper, lower, top, bottom" are used generally with respect to the orientation shown in the drawings or with respect to the positional relationship of the various components with respect to one another in the vertical, vertical or gravitational directions.
In addition, if there is a description of "first", "second", etc. in the embodiments of the present invention, the description of "first", "second", etc. is for descriptive purposes only and is not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In addition, the technical solutions of the embodiments may be combined with each other, but it is necessary to base that the technical solutions can be realized by those skilled in the art, and when the technical solutions are contradictory or cannot be realized, the combination of the technical solutions should be considered to be absent and not within the scope of protection claimed in the present invention.
Fig. 1 is a flowchart illustrating a method for recognizing a corn leaf disease image in a complex environment according to an embodiment of the present invention. In this fig. 1, the identification method may include:
in step S10, obtaining a corn leaf disease image to be identified;
in step S11, a first convolution operation is performed on the image to obtain a first feature map;
in step S12, a first pooling operation is performed on the first feature map according to formula (1) to obtain a second feature map,
f pool ＝
Max(x m,n ,x m,n+1 ,x m,n+2 ,x m+1,n ,x m+1,n+1 ,x m+1,n+2 ,x m+2,n ,x m+2,n+1 ,x m+2,n+2 )，(1)
wherein, fpool x is a pixel value in the feature map, and m and n are position information of pixels;
in step S13, a second convolution operation is performed on the second feature map to obtain a third feature map. Wherein the second convolution operation may include inputting a second feature map into the first convolution kernel to obtain a first CASF (Coordinated Attention Scale Fusion, coordinated attention mechanism scale fusion) feature; inputting the second feature map and the first CASF feature into a second convolution kernel to obtain a second CASF feature; inputting the first CASF feature and the second CASF feature into a third convolution kernel to obtain a third CASF feature; inputting the third CASF feature into a fourth convolution kernel to obtain a third feature map;
in step S14, a second pooling operation is performed on the third feature map according to formula (2) to obtain a fourth feature map,
wherein y is c For the output of the c-th channel xC (i, j), W is the width value of the input feature, and H is the height value of the input feature;
in step S15, the fourth feature map is input into the classification layer to determine the type of corn leaf disease to which the image corresponds.
In the method as shown in fig. 1, step S10 may be used to acquire a corn leaf disease image to be identified, for example, by setting an input layer to receive the corn leaf disease image to be identified. However, since the background of the input corn leaf disease image often has the characteristic of complex redundancy, the downsampling operation (the first convolution operation) needs to be performed through the step S11, the R, G, B three channels of the corn leaf disease image are converted into the 24-channel feature map, and the complex redundant background information in the feature map is removed through the first pooling operation. In particular, although the size of the convolution kernel for performing the downsampling operation may be in various forms known to those skilled in the art, it is verified through experiments performed a plurality of times by the inventor that the size of the convolution kernel may be 7*7. Similarly, the size of the first pooling operation may preferably be 3*7.
After the first pooling operation is performed in step S12, the identification method further processes the identification method in step S13 in order to make the features in the obtained second feature map more accurate. Specifically, in this step S13, the second feature map may be input into the first convolution kernel to obtain the first CASF feature; inputting the second feature map and the first CASF feature into a second convolution kernel to obtain a second CASF feature; inputting the first CASF feature and the second CASF feature into a third convolution kernel to obtain a third CASF feature; and finally, inputting the third CASF feature into a fourth convolution kernel to obtain a third feature map. The specific procedure of the second convolution operation may also be as shown in fig. 2.
In fig. 2, the first convolution kernel, the second convolution kernel, the third convolution kernel, and the fourth convolution kernel perform fusion convolution calculation on the currently input CASF feature and the CASF feature received by the previous convolution kernel, and simultaneously calculate only the currently input CASF feature and also fusion calculate the CASF feature received by the previous convolution kernel. The CASF features are parameters for representing the spot positions in the corn leaf disease image, and in the method, the first convolution kernel, the second convolution kernel, the third convolution kernel and the fourth convolution kernel are fused through the two CASF features, so that the effectiveness of identifying the corn leaf disease image is further improved compared with a conventional non-fusion mechanism, and the identification performance is further improved. Wherein, although the sizes and parameters of the first, second, third and fourth convolution kernels may be in various forms known to those skilled in the art. However, it has been experimentally demonstrated that in a preferred example of the invention, the first convolution kernel, the second convolution kernel and the third convolution kernel may each comprise three convolution kernels in series in sequence, wherein:
the first convolution kernel of the first convolution kernel has a size of 1*1, a dilation rate of 1, an activation function of a ReLU function, a second convolution kernel has a size of 3*3, a dilation rate of 1, an activation function of a ReLU function, a third convolution kernel has a size of 1*1, a dilation rate of 1, and an activation function of a ReLU function;
the first convolution kernel of the second convolution kernel has a size of 1*1, a dilation rate of 2, an activation function of a ReLU function, the second convolution kernel has a size of 3*3, a dilation rate of 2, an activation function of a ReLU function, and the third convolution kernel has a size of 1*1, a dilation rate of 2, and an activation function of a ReLU function;
the first convolution kernel of the third convolution kernel has a size of 1*1, an expansion rate of 3, an activation function of a ReLU function, a size of 3*3, an expansion rate of 3, an activation function of a ReLU function, a size of 1*1, an expansion rate of 3, and an activation function of a ReLU function;
the fourth convolution kernel includes a convolution kernel of size 3*3, step size 2, and activation function ReLU.
On the basis that the first convolution kernel, the second convolution kernel, the third convolution kernel and the fourth convolution kernel select the parameters, the number of channels in the input feature map can be increased through the first convolution kernel, so that the second convolution kernel can extract more useful features; the third convolution kernel may perform a depth-separable processing on the output of the second convolution kernel, thereby greatly reducing the number of parameters in the feature map. However, this depth-separable operation may result in loss of relationship information between the channels in the feature map, and therefore, a fourth convolution kernel may be used to create fused channel information between each channel while reducing the channel dimensions. More specifically, taking the number of channels of the feature map input to the first convolution kernel as n and the scale as m as an example. The number of channels of the feature map output by the first convolution kernel may be 2n, and the scale may be m/2. The number of channels of the feature map output by the second convolution kernel may be 4n, and the scale may be m/4. The number of channels of the feature map output by the third convolution kernel may be 8n, and the scale may be m/8. The number of channels of the feature map output by the fourth convolution kernel may be 16n.
Although the recognition rate of the corn leaf disease image can be improved to a certain extent by performing the second convolution operation once on the second feature map. However, the inventors found through experiments that the recognition rate is significantly improved by performing the second convolution operation a different number of times. Furthermore, the number of times the second convolution operation is performed is different, and the value by which the recognition rate is increased or decreased is also different. Therefore, the number of times of execution of the second convolution operation may preferably be 3 times through a large number of experiments. And after any second convolution operation, the number of channels of the feature map after operation is twice the number of channels of the feature map before operation, and the scale of the feature map after operation is half of the scale of the feature map before operation.
After the second convolution operation is performed for a plurality of times, the second pooling operation can be performed on the third feature map again before the classification operation is finally performed, so that the technical problem of overfitting is prevented.
Finally, the fourth feature map after processing is input into the classification layer through step S15 to obtain the corresponding corn leaf disease type. Specifically, the classification layer calculates the probability of the fourth feature map corresponding to each corn leaf disease type, and then selects the class with the highest probability as the final recognition result.
On the other hand, the invention also provides a system for identifying corn leaf disease images in a complex environment, and as shown in fig. 3, the system can comprise an input layer 01, a first convolution layer 02, a first pooling layer 03, a second convolution layer, a second pooling layer and a classification layer. Wherein:
the input layer 01 may be used to obtain an image of corn leaf disease to be identified. The first convolution layer 02 may be used to perform a first convolution operation on the image to obtain a first feature map. The first pooling layer 03 may be used to perform a first pooling operation on the first feature map according to equation (1) to obtain a second feature map,
f pool ＝
Max(x m,n ,x m,n+1 ,x m,n+2 ,x m+1,n ,x m+1,n+1 ,x m+1,n+2 ,x m+2,n ,x m+2,n+1 ,x m+2,n+2 )，(1)
wherein f pool X is a pixel value in the feature map, and m and n are position information of pixels; the second convolution layer 04 may be configured to perform a second convolution operation on the second feature map to obtain a third feature map. The second pooling layer 05 may be configured to perform a second pooling operation on the third feature map according to formula (2) to obtain a fourth feature map;
wherein y is c For the c-th channel x c The output of (i, j), W being the width value of the input feature and H being the height value of the input feature; the classifying layer 06 may be configured to determine a corn leaf disease type corresponding to the image according to the fourth feature map input classifying layer.
The second convolution layer 04 may have a structure as shown in fig. 2 and is three layers. Specifically, each second convolution layer 04 may include a first convolution kernel, a second convolution kernel, a third convolution kernel, and a fourth convolution kernel. The first convolution kernel may be used to generate a first CASF feature from the second feature map; the second convolution kernel may be configured to derive a second CASF feature from the second feature map and the first CASF feature; the third convolution kernel may be configured to derive a third CASF feature from the first CASF feature and the second CASF feature; the fourth convolution kernel may be used to derive a third feature map from the third CASF feature.
In fig. 2, the first convolution kernel, the second convolution kernel, the third convolution kernel, and the fourth convolution kernel perform fusion convolution calculation on the currently input CASF feature and the CASF feature received by the previous convolution kernel, and simultaneously calculate only the currently input CASF feature and also fusion calculate the CASF feature received by the previous convolution kernel. The CASF features are parameters for representing the spot positions in the corn leaf disease image, and in the method, the first convolution kernel, the second convolution kernel, the third convolution kernel and the fourth convolution kernel are fused through the two CASF features, so that the effectiveness of identifying the corn leaf disease image is further improved compared with a conventional non-fusion mechanism, and the identification performance is further improved. Wherein, although the sizes and parameters of the first, second, third and fourth convolution kernels may be in various forms known to those skilled in the art. However, it has been experimentally demonstrated that in a preferred example of the invention, the first convolution kernel, the second convolution kernel and the third convolution kernel may each comprise three convolution kernels in series in sequence, wherein:
the first convolution kernel of the first convolution kernel has a size of 1*1, a dilation rate of 1, an activation function of a ReLU function, a second convolution kernel has a size of 3*3, a dilation rate of 1, an activation function of a ReLU function, a third convolution kernel has a size of 1*1, a dilation rate of 1, and an activation function of a ReLU function;
the first convolution kernel of the second convolution kernel has a size of 1*1, a dilation rate of 2, an activation function of a ReLU function, the second convolution kernel has a size of 3*3, a dilation rate of 2, an activation function of a ReLU function, and the third convolution kernel has a size of 1*1, a dilation rate of 2, and an activation function of a ReLU function;
the first convolution kernel of the third convolution kernel has a size of 1*1, an expansion rate of 3, an activation function of a ReLU function, a size of 3*3, an expansion rate of 3, an activation function of a ReLU function, a size of 1*1, an expansion rate of 3, and an activation function of a ReLU function;
the fourth convolution kernel includes a convolution kernel of size 3*3, step size 2, and activation function ReLU.
In yet another aspect, the present invention also provides a training method for training an identification system as described in any one of the above. The steps of the training method may be as shown in fig. 4. In this fig. 4, the training method may include:
in step S20, a training set and a test set are preset;
in step S21, a training set is used to train the recognition system;
in step S22, the identification system is tested by using the test set;
in step S23, calculating an error between the classification result output by the recognition system and the standard result corresponding to the test image of the test set;
in step S24, determining whether the error is smaller than a preset error threshold;
in step S25, in the case where the error is judged to be greater than or equal to the error threshold value, the parameters of the recognition system are updated using formula (3),
l asce ＝p y 2 log(p y )l cce +(1-p y )l rce ， (3)
wherein l asce For adaptive cross entropy loss function, p y To output probability, l cce For the forward cross entropy loss, l rce Is the inverse cross entropy loss;
training the recognition system by adopting a training set again, and executing corresponding steps of the training method until the judgment error is smaller than the error threshold;
in step S20, if the error is less than the error threshold, the recognition system after training is output.
In the training method shown in fig. 4, the preset training set and test set may be obtained by a plurality of preprocessing operations such as clipping, scaling, rotation, etc. by a staff member shooting in the field through a mobile terminal or directly downloading from the internet. To ensure accuracy of the training results, the data ratio of the training set to the test set may be, for example, 7:3.
The process of step S21, step S22 and step S24 may be a training process of a neural network known to those skilled in the art, and thus, will not be described herein. In the case where the error is greater than or equal to the error threshold value in step S24, it is indicated that the current neural network (i.e., the identification system) cannot meet the accuracy requirement (error threshold value), and therefore step S25 is required to be performed to update the neural network. Although the conventional neural network parameter updating method can update the identification system provided by the invention, the neural network is difficult to train and even easy to overfit due to the fact that some difficultly-classified samples and outlier data exist in the corn leaf disease image acquired under a complex scene. Thus, in a preferred example of the invention, equation (3) may be employed to update the identification system.
In designing this equation (3), the inventors found that the cross entropy function as shown in equation (4),
wherein l cce As a cross entropy function, y i E {0,1}, i e {1, 2.,. K } is the real label, p i E {0,1}, i e {1, 2..k } is the output probability. If y is to be i And p i For applying predictive probability p y To indicate that then p y I.e. the predicted probability (output probability) corresponding to the real tag. Then the equation (4) can be rewritten as equation (5),
l cce ＝-log(p y )， (5)。
experiments have shown that one feature of the cross entropy function as shown in equation (5) is the probability p with prediction y Is reduced, l cce The function value and gradient of (c) increases dramatically. With such a premise, when a sample with a wrong classification is severely penalized, faster convergence can be achieved by the cross entropy function. However, if the dataset itself is noisy, the falsely marked data can produce abnormally large gradient values, thereby impeding the optimization process.
However, the characteristics of the inverse cross entropy function as shown in equation (6) are exactly the opposite,
wherein l rce May be an inverse cross entropy function. Due to y i Is a one-hot vector, and therefore log (0) =a can be defined directly. The inverse cross entropy function can then be expressed as equation (7),
l rce ＝-A(1-p y )， (7)。
when a= -2, the inverse cross entropy function becomes a mean absolute error loss. The convergence rate of the mean absolute error loss is very slow, which, although there is no problem that the mislabeled data caused by the cross entropy function will generate an abnormally large gradient value, thus impeding the optimization process, can cause the convergence rate to be much reduced, resulting in a significantly longer training time.
Thus, in a preferred example of the present invention, the cross entropy function and the inverse cross entropy function may be combined, i.e. equation (3) above. The formula (3) can overcome the defects of the cross entropy function and the inverse cross entropy function at the same time, and the advantages of the cross entropy function and the inverse cross entropy function are maintained, so that the training efficiency of the identification system is improved.
In yet another aspect, the present invention also provides a computer readable storage medium having stored thereon instructions which can be used to be read by a machine to cause the machine to perform a method as described in any of the above.
By the technical scheme, the identification method and the identification system for the corn leaf disease image in the complex environment realize accurate and efficient identification of the corn leaf disease image by adopting the light-weight neural network by constructing the light-weight neural network and combining with the fusion principle of the CASF characteristics, and overcome the technical defect of limitation of the application scene of the identification method caused by adopting the light-weight neural network in the prior art.
The optional embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the embodiments of the present invention are not limited to the specific details of the foregoing embodiments, and various simple modifications may be made to the technical solutions of the embodiments of the present invention within the scope of the technical concept of the embodiments of the present invention, and all the simple modifications belong to the protection scope of the embodiments of the present invention.
In addition, the specific features described in the above embodiments may be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the various possible combinations of embodiments of the invention are not described in detail.
Those skilled in the art will appreciate that all or part of the steps in implementing the methods of the embodiments described above may be implemented by a program stored in a storage medium, including instructions for causing a (which may be a single-chip microcomputer, a chip or the like) or processor (processor) to perform all or part of the steps of the methods of the embodiments described herein. 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 addition, any combination of the various embodiments of the present invention may be made between the various embodiments, and should also be regarded as disclosed in the embodiments of the present invention as long as it does not deviate from the idea of the embodiments of the present invention.
Claims (10)
1. The method for identifying the corn leaf disease image in the complex environment is characterized by comprising the following steps of:
acquiring a corn leaf disease image to be identified;
performing a first convolution operation on the image to obtain a first feature map;
performing a first pooling operation on the first feature map according to equation (1) to obtain a second feature map,
f pool ＝
Max(x m，n ，x m，n+1 ，x m，n+2 ，z m+1，n ，x m+1，n+1 ，x m+1，n+2 ，x m+2，n ，x m+2，n+1 ，x m+2，n+2 )， (1)
wherein f pool For the value after the first pooling operation, x is the pixel value in the feature map, m and n are the bits of the pixelSetting information;
performing a second convolution operation on the second feature map to obtain a third feature map, wherein the second convolution operation includes:
inputting the second feature map into a first convolution kernel to obtain a first CASF feature;
inputting the second feature map and the first CASF feature into a second convolution kernel to obtain a second CASF feature;
inputting the first and second CASF features into a third convolution kernel to obtain a third CASF feature;
inputting the third CASF feature into a fourth convolution kernel to obtain the third feature map;
performing a second pooling operation on the third feature map according to formula (2) to obtain a fourth feature map;
wherein y is c For the c-th channel x c The output of (i, j), W being the width value of the input feature and H being the height value of the input feature;
and inputting the fourth characteristic diagram into a classification layer to determine the corn leaf disease type corresponding to the image.
2. The identification method of claim 1, wherein the first convolution kernel, the second convolution kernel, and the third convolution kernel each comprise three convolution kernels in series in sequence, wherein:
the first convolution kernel of the first convolution kernel has a size of 1*1, a dilation rate of 1, an activation function of a ReLU function, a second convolution kernel has a size of 3*3, a dilation rate of 1, an activation function of a ReLU function, and a third convolution kernel has a size of 1*1, a dilation rate of 1, and an activation function of a ReLU function;
the first convolution kernel of the second convolution kernel has a size of 1*1, an expansion rate of 2, an activation function of a ReLU function, the second convolution kernel has a size of 3*3, an expansion rate of 2, an activation function of a ReLU function, and the third convolution kernel has a size of 1*1, an expansion rate of 2, and an activation function of a ReLU function;
the first convolution kernel of the third convolution kernel has a size of 1*1, an expansion rate of 3, an activation function of a ReLU function, a second convolution kernel has a size of 3*3, an expansion rate of 3, an activation function of a ReLU function, a third convolution kernel has a size of 1*1, an expansion rate of 3, and an activation function of a ReLU function;
the fourth convolution kernel includes a convolution kernel of size 3*3, step size 2, and activation function ReLU.
3. The method of identifying of claim 1, wherein performing a second convolution operation on the second signature to obtain a third signature comprises:
and executing the second convolution operation for three times, wherein after any second convolution operation, the number of channels of the feature map after operation is twice the number of channels of the feature map before operation, and the scale of the feature map after operation is half of the scale of the feature map before operation.
4. The method of claim 1, wherein inputting the fourth feature map into a classification layer to determine a corn leaf disease type for the image comprises:
inputting the fourth feature map into a classification layer to obtain the probability of each corn leaf disease type corresponding to the image;
and selecting the maize virus type with the highest probability as a final recognition result.
5. A system for identifying corn leaf disease images in a complex environment, the system comprising:
the input layer is used for acquiring a corn leaf disease image to be identified;
the first convolution layer is used for executing first convolution operation on the image to obtain a first feature map;
a first pooling layer for performing a first pooling operation on the first feature map according to formula (1) to obtain a second feature map,
f pool ＝
Max(x m，n ，x m，n+1 ，x m，n+2 ，x m+1，n ，x m+1，n+1 ，x m+1，n+2 ，x m+2，n ，x m+2，n+1 ，x m+2，n+2 )， (1)
wherein f pool X is a pixel value in the feature map, and m and n are position information of pixels;
a second convolution layer, configured to perform a second convolution operation on the second feature map to obtain a third feature map, where the second convolution layer includes:
a first convolution kernel for generating a first CASF feature from the second feature map;
a second convolution kernel, configured to obtain a second CASF feature according to the second feature map and the first CASF feature;
a third convolution kernel configured to obtain a third CASF feature according to the first CASF feature and the second CASF feature;
a fourth convolution kernel, configured to obtain the third feature map according to the third CASF feature;
a second pooling layer for performing a second pooling operation on the third feature map according to formula (2) to obtain a fourth feature map;
wherein y is c For the c-th channel x c The output of (i, j), W being the width value of the input feature and H being the height value of the input feature;
and the classifying layer is used for inputting the fourth characteristic diagram into the classifying layer to determine the corn leaf disease type corresponding to the image.
6. The identification system of claim 5, wherein the first convolution kernel, the second convolution kernel, and the third convolution kernel each comprise three convolution kernels in series in sequence, wherein:
the first convolution kernel of the first convolution kernel has a size of 1*1, a dilation rate of 1, an activation function of a ReLU function, a second convolution kernel has a size of 3*3, a dilation rate of 1, an activation function of a ReLU function, and a third convolution kernel has a size of 1*1, a dilation rate of 1, and an activation function of a ReLU function;
the first convolution kernel of the second convolution kernel has a size of 1*1, an expansion rate of 2, an activation function of a ReLU function, the second convolution kernel has a size of 3*3, an expansion rate of 2, an activation function of a ReLU function, and the third convolution kernel has a size of 1*1, an expansion rate of 2, and an activation function of a ReLU function;
the first convolution kernel of the third convolution kernel has a size of 1*1, an expansion rate of 3, an activation function of a ReLU function, a second convolution kernel has a size of 3*3, an expansion rate of 3, an activation function of a ReLU function, a third convolution kernel has a size of 1*1, an expansion rate of 3, and an activation function of a ReLU function;
the fourth convolution kernel includes a convolution kernel of size 3*3, step size 2, and activation function ReLU.
7. The identification system of claim 5, wherein the second convolution layers are three layers, each layer comprising two first convolution kernels, one second convolution kernel, and one fourth convolution kernel, wherein the number of channels of the output signature of each layer of the second convolution layers is twice the number of channels of the input signature, and the scale of the output signature is half the scale of the input signature.
8. The recognition system of claim 5, wherein the classification layer is configured to:
generating probability of each corn leaf disease type corresponding to the image according to the fourth feature map;
and selecting the maize virus type with the highest probability as a final recognition result.
9. A training method for training an identification system as claimed in any one of claims 5 to 8, comprising:
presetting a training set and a testing set;
training the recognition system by adopting the training set;
testing the identification system using the test set;
calculating errors of the classification result output by the identification system and the standard result corresponding to the test image of the test set;
judging whether the error is smaller than a preset error threshold value or not;
in the case where the error is determined to be greater than or equal to the error threshold, updating the parameters of the identification system using equation (3),
l asce ＝p y 2 log(p y )l cce +(1-p y )l rce ， (3)
wherein l asce For adaptive cross entropy loss function, p y To output probability, l cce For the forward cross entropy loss, l rce Is the inverse cross entropy loss;
training the recognition system by adopting the training set again, and executing corresponding steps of the training method until the error is judged to be smaller than the error threshold;
and outputting the recognition system after training is completed under the condition that the error is judged to be smaller than the error threshold value.
10. A computer readable storage medium storing instructions for reading by a machine to cause the machine to perform the method of any one of claims 1 to 4, 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110757395.XA CN113469064B (en) | 2021-07-05 | 2021-07-05 | Identification method and system for corn leaf disease image in complex environment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110757395.XA CN113469064B (en) | 2021-07-05 | 2021-07-05 | Identification method and system for corn leaf disease image in complex environment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113469064A CN113469064A (en) | 2021-10-01 |
| CN113469064B true CN113469064B (en) | 2024-03-29 |
Family
ID=77878035
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110757395.XA Active CN113469064B (en) | 2021-07-05 | 2021-07-05 | Identification method and system for corn leaf disease image in complex environment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113469064B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116311230B (en) * | 2023-05-17 | 2023-08-18 | 安徽大学 | Corn leaf disease identification method and device oriented to real scene |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108304812A (en) * | 2018-02-07 | 2018-07-20 | 郑州大学西亚斯国际学院 | A kind of crop disease recognition methods based on convolutional neural networks and more video images |
| CN109344699A (en) * | 2018-08-22 | 2019-02-15 | 天津科技大学 | Winter jujube disease recognition method based on depth of seam division convolutional neural networks |
| AU2020103901A4 (en) * | 2020-12-04 | 2021-02-11 | Chongqing Normal University | Image Semantic Segmentation Method Based on Deep Full Convolutional Network and Conditional Random Field |
| CN112598675A (en) * | 2020-12-25 | 2021-04-02 | 浙江科技学院 | Indoor scene semantic segmentation method based on improved full convolution neural network |
| CN112766283A (en) * | 2021-01-25 | 2021-05-07 | 西安电子科技大学 | Two-phase flow pattern identification method based on multi-scale convolution network |
-
2021
- 2021-07-05 CN CN202110757395.XA patent/CN113469064B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108304812A (en) * | 2018-02-07 | 2018-07-20 | 郑州大学西亚斯国际学院 | A kind of crop disease recognition methods based on convolutional neural networks and more video images |
| CN109344699A (en) * | 2018-08-22 | 2019-02-15 | 天津科技大学 | Winter jujube disease recognition method based on depth of seam division convolutional neural networks |
| AU2020103901A4 (en) * | 2020-12-04 | 2021-02-11 | Chongqing Normal University | Image Semantic Segmentation Method Based on Deep Full Convolutional Network and Conditional Random Field |
| CN112598675A (en) * | 2020-12-25 | 2021-04-02 | 浙江科技学院 | Indoor scene semantic segmentation method based on improved full convolution neural network |
| CN112766283A (en) * | 2021-01-25 | 2021-05-07 | 西安电子科技大学 | Two-phase flow pattern identification method based on multi-scale convolution network |
Non-Patent Citations (2)
| Title |
|---|
| 卷积神经网络在黄瓜叶部病害识别中的应用;张善文;谢泽奇;张晴晴;;江苏农业学报;20180307(第01期);全文 * |
| 基于改进VGG卷积神经网络的棉花病害识别模型;张建华;孔繁涛;吴建寨;翟治芬;韩书庆;曹姗姗;;中国农业大学学报;20181115(第11期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113469064A (en) | 2021-10-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2023138300A1 (en) | Target detection method, and moving-target tracking method using same | |
| CN111291825B (en) | Focus classification model training method, apparatus, computer device and storage medium | |
| JP2015087903A (en) | Apparatus and method for information processing | |
| CN110889863B (en) | Target tracking method based on target perception correlation filtering | |
| CN109859113B (en) | Model generation method, image enhancement method, device and computer-readable storage medium | |
| CN110675429A (en) | Long-range and short-range complementary target tracking method based on twin network and related filter | |
| CN110826609B (en) | Double-current feature fusion image identification method based on reinforcement learning | |
| CN113469064B (en) | Identification method and system for corn leaf disease image in complex environment | |
| CN114049515A (en) | Image classification method, system, electronic device and storage medium | |
| CN114639102A (en) | Cell segmentation method and device based on key point and size regression | |
| CN113111885B (en) | Dynamic resolution instance segmentation method and computer readable storage medium | |
| CN116206227B (en) | Picture examination system and method for 5G rich media information, electronic equipment and medium | |
| CN112597997A (en) | Region-of-interest determining method, image content identifying method and device | |
| CN117173568A (en) | Target detection model training method and target detection method | |
| CN114528976B (en) | Equal transformation network training method and device, electronic equipment and storage medium | |
| CN115862119A (en) | Human face age estimation method and device based on attention mechanism | |
| Jayageetha et al. | Medical image quality assessment using CSO based deep neural network | |
| CN115512207A (en) | Single-stage target detection method based on multipath feature fusion and high-order loss sensing sampling | |
| CN113284122A (en) | Method and device for detecting roll paper packaging defects based on deep learning and storage medium | |
| CN111144422A (en) | Positioning identification method and system for aircraft component | |
| CN117173538A (en) | Model training method and device, nonvolatile storage medium and electronic equipment | |
| CN116091867B (en) | Model training and image recognition method, device, equipment and storage medium | |
| CN117094978A (en) | Multi-scale gradient selection self-adaptive pavement defect detection method | |
| CN116245619B (en) | Commodity vector embedding method, commodity similarity evaluation method and commodity display method | |
| CN114067370B (en) | Neck shielding detection method and device, electronic equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |