CN111382676A - Sand image classification method based on attention mechanism - Google Patents
Sand image classification method based on attention mechanism Download PDFInfo
- Publication number
- CN111382676A CN111382676A CN202010114805.4A CN202010114805A CN111382676A CN 111382676 A CN111382676 A CN 111382676A CN 202010114805 A CN202010114805 A CN 202010114805A CN 111382676 A CN111382676 A CN 111382676A
- Authority
- CN
- China
- Prior art keywords
- sand
- image
- sand grain
- attention mechanism
- grain image
- 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.)
- Granted
Links
- 239000004576 sand Substances 0.000 title claims abstract description 107
- 238000000034 method Methods 0.000 title claims abstract description 34
- 230000007246 mechanism Effects 0.000 title claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 34
- 238000013145 classification model Methods 0.000 claims abstract description 14
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 239000010453 quartz Substances 0.000 claims description 13
- 239000011435 rock Substances 0.000 claims description 13
- VYPSYNLAJGMNEJ-UHFFFAOYSA-N silicon dioxide Inorganic materials O=[Si]=O VYPSYNLAJGMNEJ-UHFFFAOYSA-N 0.000 claims description 13
- -1 shale Substances 0.000 claims description 6
- 239000013078 crystal Substances 0.000 claims description 5
- 229910052655 plagioclase feldspar Inorganic materials 0.000 claims description 5
- 239000010454 slate Substances 0.000 claims description 4
- 229910052650 alkali feldspar Inorganic materials 0.000 claims description 2
- 238000013527 convolutional neural network Methods 0.000 abstract description 12
- 238000012360 testing method Methods 0.000 abstract description 4
- 238000010606 normalization Methods 0.000 abstract description 2
- 239000010410 layer Substances 0.000 description 42
- 238000011176 pooling Methods 0.000 description 19
- 230000006870 function Effects 0.000 description 17
- 238000010586 diagram Methods 0.000 description 8
- 238000013461 design Methods 0.000 description 5
- 239000010433 feldspar Substances 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000011160 research Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 239000000284 extract Substances 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical compound [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 229910052500 inorganic mineral Inorganic materials 0.000 description 2
- 239000011707 mineral Substances 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 239000008186 active pharmaceutical agent Substances 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 238000003776 cleavage reaction Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000003628 erosive effect Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 229910001385 heavy metal Inorganic materials 0.000 description 1
- 239000011229 interlayer Substances 0.000 description 1
- 229910052742 iron Inorganic materials 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000003014 reinforcing effect Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 230000007017 scission Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
- G06V20/698—Matching; Classification
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
The invention discloses a sand grain image classification method based on an attention mechanism, which comprises the following steps: 1) collecting a certain number of sand grain images marked with categories to construct a training data set; 2) carrying out scaling and normalization preprocessing on the sand images in the training data set to enable the image data to meet the network input requirement; 3) designing a convolutional network structure based on an attention mechanism, wherein the convolutional network structure comprises a basic network framework and an attention module; 4) defining a loss function, training an automatic sand image classification model, and obtaining an end-to-end multi-class automatic sand image classification model; 5) inputting the preprocessed test sand grain image into a model for prediction, and outputting the category of the sand grain image. Compared with the traditional sand grain image classification method based on feature engineering, the method provided by the invention is different in that a convolutional neural network and an attention mechanism are introduced, so that the classification accuracy of the sand grain image is improved, and the classification speed is accelerated.
Description
Technical Field
The invention belongs to the technical field of sand image classification, and particularly relates to a sand image classification method based on an attention mechanism.
Background
River sand is a recording carrier of geological information such as material source minerals and rocks. Through river sand research, basic information such as weathering, carrying, erosion, structure and the like of different geologic bodies and different weather zones can be obtained, and the method is an important way and an effective means for understanding the modern structure and landform evolution. In addition, the river sand research can help researchers to explore the evolution of the water system of the great river and has important significance for understanding the evolution of the sandstone and the sedimentary basin in the geological history.
Classification of microscopically thin slice images of sand is an important task in geological research. Through accurate identification and identification of the lithology of the sand grains, geologists can be helped to obtain distribution information of various sand grain components in the river quicksand. The traditional method for identifying the lithology of the sand grains is to manually identify sand grain images. Due to the fact that geological professional knowledge is needed and the workload is huge, the problems of low accuracy, long consumed time, easiness in being influenced by subjective factors and the like exist.
Convolutional neural networks are a class of feed-forward neural networks that contain convolutional computations and have a deep structure. The convolutional neural network is constructed by imitating a biological visual mechanism, and the convolutional kernel parameter sharing in an implicit layer and the sparsity of interlayer connection enable the convolutional neural network to learn the image characteristics with smaller calculation amount. When the visual attention mechanism imitates the human brain to process visual signals, more attention resources are put into a target area needing important attention, and the neural network has the capacity of focusing on the effective characteristic subset of the neural network through the mechanism. At present, the automatic classification task of the sand grain images is mostly realized by adopting a mode based on feature engineering, image information is represented by artificial design and effective features, and the classification task is completed by using a machine learning algorithm. However, the method is complicated in process and low in accuracy, and a more effective deep learning method based on the convolutional neural network is not well applied.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to provide an attention mechanism-based sand image classification method, which automatically extracts sand image characteristics by designing a convolution network structure, particularly introduces the attention mechanism to extract key information in the sand image, defines a loss function to process the problem of unbalanced quantity of various samples, and effectively completes classification tasks of various sand grains such as single crystal quartz, polycrystalline quartz, plagioclase feldspar, alkali feldspar, volcanic rock, invaded rock, shale, slate, gneiss and the like.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
the invention discloses a sand grain image classification method based on an attention mechanism, which comprises the following steps of:
1) collecting the marked sand grain images to construct a training set;
2) preprocessing the sand grain image;
3) designing a convolution network structure based on an attention mechanism;
4) training a sand grain image automatic classification model;
5) and predicting the category of the sand image by using the automatic classification model of the sand image.
Further, the step 1) specifically includes: aiming at different river drift sand categories, the method comprises the following steps: the method comprises the steps of collecting single crystal quartz, polycrystalline quartz, plagioclase, alkaline feldspar, volcanic rock, invaded rock, shale, slate and gneiss, collecting sand grain images of each category, and constructing a training set based on the images, wherein the training set comprises a sand grain image set and a corresponding sand grain category label set.
Further, the step 2) specifically includes: and (4) enabling each sand grain image to have a uniform tensor format and being used as input of a convolution network model.
Further, the convolutional network structure in step 3) includes: an underlying network framework and an attention module.
Further, the step 4) specifically includes: based on the training set and the loss function definition, the marked sand grain image is utilized to complete convolution network training based on an attention mechanism, and an end-to-end sand grain image multi-class automatic classification model is obtained.
Further, the step 5) specifically includes: inputting a sand image obtained by taking a picture under a microscope, namely new sample data, predicting the new sample data by using the trained model, and outputting the category and the confidence coefficient of the sand image.
The invention has the beneficial effects that:
the method can solve the problem of unbalanced quantity of the sand grains, and improve the attention of the model to important detail characteristics of the sand grain image, thereby effectively improving the precision of fine classification of the sand grains; the method has high network training speed, can quickly finish sand classification, is suitable for automatic classification of massive sand images, has good expansibility, robustness and practicability, and completely meets the statistical work requirement of geological research on the quantity distribution of various sands in river sand.
Drawings
FIG. 1 is an overall block diagram of the method of the present invention.
Fig. 2 is a class diagram of sand classification.
Fig. 3 is a general block diagram of a convolutional network of the present invention.
FIG. 4 is a structural layout diagram of an attention module.
FIG. 5 is a schematic diagram of the results output by the attention module.
Detailed Description
In order to facilitate understanding of those skilled in the art, the present invention will be further described with reference to the following examples and drawings, which are not intended to limit the present invention.
Referring to fig. 1, the sand image classification method based on attention mechanism of the present invention includes the following steps:
1) collecting the marked sand grain images to construct a training set; aiming at different river sand categories including single crystal quartz, polycrystalline quartz, plagioclase, alkaline feldspar, volcanic rock, invaded rock, shale, slate, gneiss and the like, sand grain images of each category are collected, and a training set is constructed based on the images, wherein the training set comprises a sand grain image set and a corresponding sand grain category label set.
2) Preprocessing the sand grain image; and (4) enabling each sand grain image to have a uniform tensor format and being used as input of a convolution network model.
3) A convolutional network structure based on an attention mechanism is designed, and comprises a basic network framework and an attention module.
4) Training an automatic sand image classification model, and completing convolution network training based on an attention mechanism by using the marked sand image based on a training set and loss function definition to obtain an end-to-end sand image multi-class automatic classification model.
5) And predicting the category of the sand image by using the automatic classification model of the sand image.
The general flow of the invention is divided into two stages: a training phase and a testing phase. The main task of the training stage is to train by using the marked data to obtain a multi-class sand image classification model based on a convolutional neural network and an attention mechanism, the main work of the testing stage is to input a sand image obtained by photographing under a microscope, namely new sample data, predict the new sample data by using the trained model, and output the class and the confidence coefficient of the sand image.
The sand grain image preprocessing comprises the steps of taking a sand grain microscopic image as a classification sample, enabling the image to contain a sand grain, using a bilinear interpolation algorithm to scale an original sand grain image to 256 × 256 pixels, and performing normalization processing on pixel values.
The basic network frame is designed into a two-section structure:
the first section is a feature extraction section, and the image feature extraction work is completed by the combination of convolution and pooling; the method specifically comprises the following steps: convolutional layer 1.1, convolutional layer 1.2, convolutional layer 1.3, pooling layer 1, convolutional layer 2.1, convolutional layer 2.2, convolutional layer 2.3, pooling layer 2, convolutional layer 3.1, convolutional layer 3.2, convolutional layer 3.3, pooling layer 3, convolutional layer 4.1, convolutional layer 4.2, convolutional layer 4.3, pooling layer 4, convolutional layer 5.1, convolutional layer 5.2, convolutional layer 5.3, pooling layer 5; after each convolution operation is finished, the linear rectification activation function is used for processing the output, and the nonlinear expression capability of the model is enhanced; after each pooling operation is finished, batch standardization processing is carried out on the output result, so that the distribution of the output result follows standard normal distribution with the mean value of 0 and the variance of 1, and the network convergence speed is accelerated;
the second segment is a feature mapping segment, the automatic image features are mapped to the sand category, and the confidence coefficient is calculated; the method specifically comprises the following steps: the full connection layer 1, the full connection layer 2 and the output layer; on one hand, the high-dimensional image feature representation learned in the previous segment of network is mapped to a sample mark space through the combination of a full connection layer and an output layer; on the other hand, the number of network parameters is increased through the full connection layer, and the complexity of the model is improved, so that the learning capability of the model is enhanced.
The infrastructure network integrity structure and parameter settings are shown in table 1 below:
TABLE 1
The design steps of the attention module are as follows: according to the characteristics of the sand grain images, a hierarchical network module based on an image attention mechanism is designed, and attention modules which are designed in the same way but have different scales are respectively embedded into three positions of a basic network; the input and output sizes of the three attention modules embedded in the network are shown in the following table 2, and fig. 3 is a specific internal structure of the attention module;
TABLE 2
For each attention module, inputting a feature tensor and outputting a corresponding weight tensor; the weight tensor is multiplied by the output tensor of the basic network pooling layer in an opposite position mode to obtain an adjusted characteristic tensor;
referring to FIG. 4, the attention module comprises a two-part structure: the space attention structure and the channel attention structure enhance the network learning ability by combining the two attention structures, thereby achieving the aim of improving the classification accuracy;
the spatial attention structure is used for enhancing the key point information on the image spatial position, and comprises: maximum pooling layer, average pooling layer, and convolutional layer using 1 x 1 convolution kernel, the calculation formula for each layer is as follows:
maximum pooling layer:
SMP(x，y)＝Max(∪(m，n)S(x+m，y+n)) (1)
average pooling layer:
SAP(x，y)＝Mean(∑(m，n)SMP(x+m，y+n)) (2)
and (3) rolling layers:
Conv(x，y)＝∑cS[MP，AP](x，y)×Kc(3)
wherein, x is more than 0, M is more than y, M is more than 0, n is more than 2, C is more than 0 and is more than C, S represents the characteristic diagram tensor of the input attention module, and the input size is M × C; sMPIs an intermediate result of the feature graph through the maximum pooling layer, SAPIs the result of the feature map passing through the average pooling layer, S[MP，AP]Representing the result obtained by splicing the tensors obtained by the maximum pooling and the average pooling; conv is the eigen map matrix obtained after convolution, and K is the convolution kernel of 1 × 1 used for convolution.
The channel attention structure is used for reinforcing the influence of the key characteristic channel on the sand grain image classification result; comprises the following steps: a global average pooling layer and a channel weight generation layer; after the feature map component on each channel is input into the global average pooling layer, a scalar z is obtainedcThe calculation formula is as follows:
channel weight generation layer adds α a weight to each scalarc(initialization value is 0.1) and then an output result is obtained:
wc＝zc×αc(5)
wherein, 0 < C < C, S represents the feature map tensor of the input attention module, and the input size is M × C;
and finally, combining the outputs of the space attention structure and the channel attention structure, and enabling each weight value to fall between [0 and 1] by using a Sigmoid function to obtain an output attention tensor AM, wherein a specific calculation formula is as follows:
taking the attention module 1 as an example, when the feature map with the size of 256 × 64 is input into the attention module, 1 weight map with the size of 64 × 64 is obtained. And (3) carrying out counterpoint multiplication on the weight graph and the feature graph obtained after the pooling layer 1 to obtain the feature graph adjusted by the attention module 1, and inputting the feature graph into the network for further processing. The network structure after the attention module is embedded is shown in fig. 2.
Wherein, the loss function used for training the sand automatic classification model in the step 4) is defined as follows: aiming at the problem that the distribution of different types of sand grains in river sand has larger imbalance, in order to enhance the generalization performance of a model, a category-related cross entropy loss function Crloss is defined based on cross entropy loss, and the formula is as follows:
where M is the number of samples taken from the dataset at a time for training; n is the number of categories in the dataset; y ism,nIs the true mark corresponding to the sample m, when n is the category to which the sample m belongs, y m,n1, otherwise ym,n＝0；pm,nRepresenting the predicted probability value on the category n output by the m-th batch input network training, p 'in the formula'm,nThe definition is as follows:
αnfor training set belonging to class nThe reciprocal of the difference of the number proportion of the sand images in all the sand images is calculated according to the following formula:
wherein M isnRepresenting the total number of the sand grains with the category of n in the training sample, and K representing the total number of the categories in the sample.
And after the network structure and the loss function are designed, inputting the data set into the constructed convolutional neural network for training to obtain a convolutional neural network classification model. A random gradient descent method is adopted in the training process; random gradient descent is a common neural network optimization method, which randomly extracts a batch of samples from all training samples each time, inputs the samples into a network, obtains output after forward propagation, calculates a loss function according to an output result, updates each parameter value in the network through a backward propagation algorithm, then extracts a group of the parameters, updates the network parameters again, and repeats the steps. Different from the conventional convolutional neural network training, the network structure of the invention contains an additional embedded branch structure, so that the basic parameters and the attention module parameters of the network need to be optimized simultaneously by means of a back propagation algorithm, and the loss function is continuously reduced. And obtaining the optimized network model after a plurality of iterations.
And (3) entering a testing stage after the model training is finished, preprocessing one sand grain image input into the model, then transmitting image data in a forward direction in a network, and finally outputting the class and the corresponding probability of the sand grain predicted by the model.
Experiments prove that the attention plays a great role in improving the classification accuracy of the sand images. Taking the first attention module as an example, after the data passes through the attention module, a weight value tensor representing the importance degree of each pixel point in the characteristic diagram is obtained. The tensor represents the correction of the attention module on the characteristic diagram, and the larger the characteristic value is, the larger the influence of the point on the classification is, which shows that the module strengthens the attention of the depth network on the hot point of the image, namely the important characteristic point. As shown in fig. 5, hot spots characterizing the quartz class in the original image are effectively emphasized. It can be observed that the three attention modules focus on different hot spots, the attention module 1 focuses more on detailed information such as textures and cleavage on the sand image, and the attention modules 2 and 3 focuses more on detailed information such as sand shape in the image, so that the hierarchical attention module design effectively improves the efficiency of extracting the sand image features by the network.
Compared with the conventional classification cross entropy loss function, the loss function CrLoss designed by aiming at the model in the invention firstly considers the unbalance of the distribution of each class in the samples, improves the influence of a small number of samples on the loss function, and thus enhances the attention of the model to the small number of classes; and secondly, the CrLoss improves the attention of the model to uncertain samples, increases the proportion of samples with low prediction confidence coefficient in the loss function, and inhibits the influence of samples with high prediction confidence coefficient in the loss function, so that the model can learn how to classify the difficult cases in the samples continuously.
The invention improves and designs a convolutional neural network with hierarchical attention module branches on the basis of a common convolutional neural network structure. Meanwhile, the unbalance problem of the quantity of the sand grains is considered, the traditional cross entropy loss function is not used, and a brand new class correlation loss function CrLoss is provided. Experiments prove that the attention mechanism is introduced to improve the attention of the model to important detail characteristics of the sand grain images, and the CrLoss design reduces the influence of the class imbalance problem on the classification accuracy to a certain extent. The classification accuracy of the quartz, the feldspar and the rock debris in three general categories reaches 93%, and under the more subdivided classification standard, the classification accuracy of the quartz, the feldspar and the rock debris specifically comprises single crystal quartz, polycrystalline quartz, plagioclase feldspar, alkaline feldspar, volcanic rock, invaded rock, shale, iron particles and other heavy metal minerals, and reaches 82%. Compared with a basic convolutional neural network model, the method has the advantages that the classification accuracy and recall rate are greatly improved, the optimization effect is more obvious in a few categories, and the statistical work requirement on the quantity distribution of various sand grains in river sand in the geological research process is completely met.
While the invention has been described in terms of its preferred embodiments, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention.
Claims (6)
1. A sand grain image classification method based on an attention mechanism is characterized by comprising the following steps:
1) collecting the marked sand grain images to construct a training set;
2) preprocessing the sand grain image;
3) designing a convolution network structure based on an attention mechanism;
4) training a sand grain image automatic classification model;
5) and predicting the category of the sand image by using the automatic classification model of the sand image.
2. The sand grain image classification method based on the attention mechanism as claimed in claim 1, wherein the step 1) specifically comprises: aiming at different river drift sand categories, the method comprises the following steps: single crystal quartz, polycrystalline quartz, plagioclase feldspar, alkali feldspar, volcanic rock, invaded rock, shale, slate, gneiss; and collecting the sand images of each category, and constructing a training set based on the images, wherein the training set comprises a sand image set and a corresponding sand category label set.
3. The sand grain image classification method based on the attention mechanism as claimed in claim 1, wherein the step 2) specifically comprises: and (4) enabling each sand grain image to have a uniform tensor format and being used as input of a convolution network model.
4. The sand grain image classification method based on the attention mechanism as claimed in claim 1, wherein the convolution network in the step 3) comprises: an underlying network framework and an attention module.
5. The sand grain image classification method based on the attention mechanism as claimed in claim 1, wherein the step 4) specifically comprises: based on the training set and the loss function definition, the marked sand grain image is utilized to complete convolution network training based on an attention mechanism, and an end-to-end sand grain image multi-class automatic classification model is obtained.
6. The sand grain image classification method based on the attention mechanism as claimed in claim 1, wherein the step 5) specifically comprises: inputting a sand image obtained by taking a picture under a microscope, namely new sample data, predicting the new sample data by using the trained model, and outputting the category and the confidence coefficient of the sand image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010114805.4A CN111382676B (en) | 2020-02-25 | 2020-02-25 | Sand grain image classification method based on attention mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010114805.4A CN111382676B (en) | 2020-02-25 | 2020-02-25 | Sand grain image classification method based on attention mechanism |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111382676A true CN111382676A (en) | 2020-07-07 |
| CN111382676B CN111382676B (en) | 2023-12-22 |
Family
ID=71217001
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010114805.4A Active CN111382676B (en) | 2020-02-25 | 2020-02-25 | Sand grain image classification method based on attention mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111382676B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111881811A (en) * | 2020-07-24 | 2020-11-03 | 浙江大学 | Self-accretion attention pooling mechanism-based rock classification method for tunneling equipment |
| CN111930982A (en) * | 2020-07-20 | 2020-11-13 | 南京南瑞信息通信科技有限公司 | Intelligent labeling method for power grid images |
| CN112801942A (en) * | 2020-12-31 | 2021-05-14 | 广西慧云信息技术有限公司 | Citrus huanglongbing image identification method based on attention mechanism |
| CN113435409A (en) * | 2021-07-23 | 2021-09-24 | 北京地平线信息技术有限公司 | Training method and device of image recognition model, storage medium and electronic equipment |
| CN113469198A (en) * | 2021-06-30 | 2021-10-01 | 南京航空航天大学 | Image classification method based on improved VGG convolutional neural network model |
| CN113537235A (en) * | 2021-02-08 | 2021-10-22 | 中国石油化工股份有限公司 | Rock identification method, system, device, terminal and readable storage medium |
| CN113610164A (en) * | 2021-08-10 | 2021-11-05 | 北京邮电大学 | Fine-grained image recognition method and system based on attention balance |
| CN115081451A (en) * | 2022-06-30 | 2022-09-20 | 中国电信股份有限公司 | Entity identification method and device, electronic equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106557758A (en) * | 2016-11-25 | 2017-04-05 | 南京大学 | A kind of multiple target automatic identification method of grains of sand micro-image |
| CN110516596A (en) * | 2019-08-27 | 2019-11-29 | 西安电子科技大学 | Empty spectrum attention hyperspectral image classification method based on Octave convolution |
| CN110598029A (en) * | 2019-09-06 | 2019-12-20 | 西安电子科技大学 | Fine-grained image classification method based on attention transfer mechanism |
-
2020
- 2020-02-25 CN CN202010114805.4A patent/CN111382676B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106557758A (en) * | 2016-11-25 | 2017-04-05 | 南京大学 | A kind of multiple target automatic identification method of grains of sand micro-image |
| CN110516596A (en) * | 2019-08-27 | 2019-11-29 | 西安电子科技大学 | Empty spectrum attention hyperspectral image classification method based on Octave convolution |
| CN110598029A (en) * | 2019-09-06 | 2019-12-20 | 西安电子科技大学 | Fine-grained image classification method based on attention transfer mechanism |
Non-Patent Citations (2)
| Title |
|---|
| HUIZHEN HAO, RONGHUA GUO, QING GU, XIUMIAN HU: "Machine learning application to automatically classify heavy minerals in river sand by using SEM/EDS data", pages 105899 * |
| NA LI, HUIZHEN HAO, ZHIWEI JIANG, FENG JIANG, RONGHUA GUO, QING GU, XIUMIAN HU: "A multi-task multi-class learning method for automatic identification of heavy minerals from river sand", pages 104403 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111930982A (en) * | 2020-07-20 | 2020-11-13 | 南京南瑞信息通信科技有限公司 | Intelligent labeling method for power grid images |
| CN111881811A (en) * | 2020-07-24 | 2020-11-03 | 浙江大学 | Self-accretion attention pooling mechanism-based rock classification method for tunneling equipment |
| CN111881811B (en) * | 2020-07-24 | 2022-04-26 | 浙江大学 | Self-accretion attention pooling mechanism-based rock classification method for tunneling equipment |
| CN112801942A (en) * | 2020-12-31 | 2021-05-14 | 广西慧云信息技术有限公司 | Citrus huanglongbing image identification method based on attention mechanism |
| CN112801942B (en) * | 2020-12-31 | 2023-10-13 | 广西慧云信息技术有限公司 | Citrus yellow dragon disease image identification method based on attention mechanism |
| CN113537235A (en) * | 2021-02-08 | 2021-10-22 | 中国石油化工股份有限公司 | Rock identification method, system, device, terminal and readable storage medium |
| CN113469198A (en) * | 2021-06-30 | 2021-10-01 | 南京航空航天大学 | Image classification method based on improved VGG convolutional neural network model |
| CN113435409A (en) * | 2021-07-23 | 2021-09-24 | 北京地平线信息技术有限公司 | Training method and device of image recognition model, storage medium and electronic equipment |
| CN113610164A (en) * | 2021-08-10 | 2021-11-05 | 北京邮电大学 | Fine-grained image recognition method and system based on attention balance |
| CN113610164B (en) * | 2021-08-10 | 2023-12-22 | 北京邮电大学 | Fine granularity image recognition method and system based on attention balance |
| CN115081451A (en) * | 2022-06-30 | 2022-09-20 | 中国电信股份有限公司 | Entity identification method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111382676B (en) | 2023-12-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111382676B (en) | Sand grain image classification method based on attention mechanism | |
| CN111368896B (en) | Hyperspectral remote sensing image classification method based on dense residual three-dimensional convolutional neural network | |
| WO2022160771A1 (en) | Method for classifying hyperspectral images on basis of adaptive multi-scale feature extraction model | |
| CN108038445B (en) | SAR automatic target identification method based on multi-view deep learning framework | |
| CN111695467B (en) | Spatial spectrum full convolution hyperspectral image classification method based on super-pixel sample expansion | |
| CN112347970B (en) | Remote sensing image ground object identification method based on graph convolution neural network | |
| CN114092832B (en) | High-resolution remote sensing image classification method based on parallel hybrid convolutional network | |
| CN109871875B (en) | Building change detection method based on deep learning | |
| Tan et al. | Vehicle detection in high resolution satellite remote sensing images based on deep learning | |
| CN111401426B (en) | Small sample hyperspectral image classification method based on pseudo label learning | |
| CN112017178A (en) | Remote sensing image region change detection method based on double-temporal difference image method | |
| Huang et al. | Research on urban modern architectural art based on artificial intelligence and GIS image recognition system | |
| CN111639878A (en) | Landslide risk prediction method and system based on knowledge graph construction | |
| CN114417693A (en) | Ocean three-dimensional temperature field inversion method based on deep learning | |
| CN108090507A (en) | A kind of medical imaging textural characteristics processing method based on integrated approach | |
| CN113095479A (en) | Method for extracting ice-below-layer structure based on multi-scale attention mechanism | |
| CN115497006B (en) | Urban remote sensing image change depth monitoring method and system based on dynamic mixing strategy | |
| CN115034478B (en) | Traffic flow prediction method based on field self-adaption and knowledge migration | |
| CN115620068A (en) | Rock lithology automatic identification and classification method under deep learning mode | |
| CN115984689A (en) | Multi-scale earth surface complexity feature extraction and land utilization segmentation method | |
| CN109993050A (en) | Synthetic aperture radar image identification method | |
| CN114219049A (en) | Fine-grained pencil and stone image classification method and device based on hierarchy constraint | |
| CN113222071A (en) | Rock classification method based on rock slice microscopic image deep learning | |
| Tian et al. | A lightweight multitask learning model with adaptive loss balance for tropical cyclone intensity and size estimation | |
| CN116563313B (en) | Remote sensing image soybean planting region segmentation method based on gating and attention fusion |
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 |