CN112990106A - Underwater object detection method and device, computer equipment and storage medium - Google Patents
Underwater object detection method and device, computer equipment and storage medium Download PDFInfo
- Publication number
- CN112990106A CN112990106A CN202110418252.6A CN202110418252A CN112990106A CN 112990106 A CN112990106 A CN 112990106A CN 202110418252 A CN202110418252 A CN 202110418252A CN 112990106 A CN112990106 A CN 112990106A
- Authority
- CN
- China
- Prior art keywords
- network
- depth estimation
- classification
- inputting
- training
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- 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
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/30—Assessment of water resources
Abstract
The method comprises the steps of obtaining a hyperspectral water body image, constructing an underwater object detection network, wherein the underwater object detection network comprises a joint anomaly detector, a depth estimation network and a classification network, extracting target pixel points with high confidence level from an input hyperspectral water body image by using the joint anomaly detector, and inhibiting the influence of background pixels on a detection result while obtaining a separation result with strong robustness; and then training the depth estimation network by using the extracted abnormal pixel point set to obtain depth prediction information, constructing a classification network training set by using the prediction result of the depth estimation network and a down-sampling method, training the classification network by using the classification network training set to obtain a trained classification network, processing the input image by using the trained classification network, and classifying the result. The method can detect and identify the specific underwater object in any scene.
Description
Technical Field
The present application relates to the field of underwater object detection technologies, and in particular, to an underwater object detection method, apparatus, computer device, and storage medium.
Background
In the field of remote sensing oceanography, a given sea area needs to be observed and monitored by means of relevant data acquired by an unmanned platform, wherein detection and identification of a specific object are important rings for realizing ocean observation. However, since the unmanned platform usually works at high altitude, the collected image data has the characteristics of large imaging field of view, low spatial resolution and the like. Therefore, the remote sensing image data cannot well complete the detection and identification tasks by using the correlation algorithm of the optical characteristics of the target due to the problems of key texture information loss and the like. However, if the underwater related objects cannot be detected and identified, the key information in the sea scene cannot be extracted, so that technical support cannot be provided for understanding the related research work of the sea scene, and finally the development process of the whole sea exploration research field is slowed down.
When the underwater object is detected and identified through the hyperspectral image, the related information of the object characteristic is lost due to the interference of the external water environment. The existing method mainly detects and identifies a specific underwater object by means of the prior spectral information of the object through the technical process of 'water body attribute inversion, object depth estimation and underwater object identification'. In a key technical link, a water body attribute inversion correlation method can achieve a high effect, but correlation research on object depth estimation also exists in a theoretical stage. Due to the fact that the action mechanism between the fingerprint information of the object and the background information of the water body is quite complex, and the prior knowledge about the depth information of the object is relatively deficient, the existing depth estimation algorithm is low in general precision, and the nearest detection result is influenced.
Disclosure of Invention
In view of the above, it is necessary to provide an underwater object detection method, apparatus, computer device and storage medium for solving the above technical problems.
A method of underwater object detection, the method comprising:
and acquiring a hyperspectral water body image, and processing the hyperspectral water body image to obtain a spectral curve.
And constructing an underwater object detection network, wherein the underwater object detection network comprises a joint anomaly detector, a depth estimation network and a classification network.
And inputting the spectrum curve into a joint anomaly detector to obtain a fusion anomaly detection result.
And taking the abnormal pixel point set with higher confidence in the fusion abnormal detection result as a depth estimation network training sample.
And training the deep estimation network by using the deep estimation network training sample to obtain the trained deep estimation network.
And inputting the spectral curve into the trained depth estimation network to obtain an object depth estimation result, and performing down-sampling according to the object depth estimation result to obtain a classification training sample.
And training the classification network by using the classification training sample to obtain the trained classification network.
Inputting the spectral curve into the trained classification network to obtain a classification result; taking all target pixel points in the classification result as the depth estimation network training set, and performing next round of training on the depth estimation network and the classification network until a preset condition is reached to obtain an underwater object detection model; the preset condition is that parameters of the depth estimation network and the classification network are converged or reach a preset training round number.
And acquiring a hyperspectral water body image to be detected, processing the hyperspectral water body image to obtain a spectral curve to be detected, and inputting the spectral curve to be detected into the underwater object detection model to obtain an underwater object detection result.
In one embodiment, the joint anomaly detector comprises a plurality of different anomaly detection modules and a decision fusion module; the plurality of different anomaly detection modules includes: RX abnormity detection module, LRX abnormity detection module, CRD abnormity detection module and CBAD abnormity detection module. Inputting the spectrum curve into a joint anomaly detector to obtain a fusion anomaly detection result; the method comprises the following steps:
and inputting the spectrum curve into the RX abnormity detection module to obtain an RX abnormity detection result.
And inputting the spectral curve into the LRX anomaly detection module to obtain an LRX anomaly detection result.
And inputting the spectrum curve into the CRD abnormity detection module to obtain a CRD abnormity detection result.
And inputting the spectrum curve into the CBAD abnormity detection module to obtain a CBAD abnormity detection result.
And inputting the RX abnormity detection result, the LRX abnormity detection result, the CRD abnormity detection result and the CBAD abnormity detection result into the decision fusion module, and performing decision fusion by adopting an ensemble learning algorithm to obtain a fusion abnormity detection result.
In one embodiment, the depth estimation network comprises a first convolutional neural network, a fully-connected network and a water body model; training the deep estimation network by using the deep estimation network training sample to obtain a trained deep estimation network, wherein the training comprises the following steps:
and inputting the depth estimation network training sample into the first convolution neural network to obtain a nonlinear convolution spectral feature.
And inputting the nonlinear convolution spectral features into the full-connection network to obtain object depth estimation.
And inputting the object depth estimation into the water body model to obtain a reconstructed spectrum curve.
And taking the difference value between the reconstructed spectral curve and the input spectral curve as a target function, adding object depth information numerical value constraint, and performing reverse training on the depth estimation network by utilizing a depth network training and gradient descent algorithm until the weight parameter of the depth estimation network is converged or reaches a preset training round number to obtain the trained depth estimation network.
In one embodiment, the first convolutional neural network comprises a one-dimensional convolutional layer, a batch normalization layer and a pooling layer; the fully connected network includes a flatten layer and a fully connected layer. Inputting the depth estimation network training sample into the first convolutional neural network to obtain a nonlinear convolutional spectrum feature, including:
inputting the depth estimation network training sample into a first convolution neural network, and performing feature extraction on the depth estimation network training sample by using a one-dimensional convolution layer with a receptive field size of 1 multiplied by 3, a step length of 1 and zero padding number of 1 to obtain convolution features.
And carrying out batch normalization processing on the convolution characteristics to obtain normalized convolution characteristics.
And utilizing the pooling layer to carry out down-sampling on the normalized convolution characteristics to obtain nonlinear convolution spectral characteristics.
In one embodiment, the fully connected network includes a flatten layer and a fully connected layer. Inputting the nonlinear convolution spectral features into the fully-connected network to obtain an object depth estimate, comprising:
and inputting the nonlinear convolution spectral features into a flatten layer to obtain flattened spectral features.
And inputting the flattened spectral features into the full-connection layer for transformation to obtain object depth estimation.
In one embodiment, inputting the spectral curve into the trained depth estimation network to obtain an object depth estimation result, and obtaining a classification training sample by using a down-sampling method according to the object depth estimation result, including:
inputting the spectral curve into the trained depth estimation network to obtain the corresponding depth of each pixel point of the hyperspectral water body image, and classifying the image according to depth information to obtain a target pixel set and a background pixel set.
And obtaining the same number of target pixel points and background pixel points by utilizing a downsampling mode and constructing a classification training sample according to the target pixel points and the background pixel points.
In one embodiment, training the classification network by using the classification training sample to obtain a trained classification network includes:
and inputting the classified training samples into a second convolutional neural network to obtain nonlinear convolutional spectral characteristics.
And inputting the nonlinear convolution spectral features into the full-connection network to obtain pixel category estimation.
And carrying out reverse training on the classification network by adopting a gradient descent algorithm according to the pixel class estimation and the classification training sample until the classification network is converged to obtain the trained classification network.
An underwater object detection device, the device comprising:
the image data acquisition module is used for acquiring a hyperspectral water body image and processing the hyperspectral water body image to obtain a spectral curve.
The network construction module is used for constructing an underwater object detection network, and the underwater object detection network comprises a joint anomaly detector, a depth estimation network and a classification network.
The training sample construction module is used for inputting the spectral curve into a joint anomaly detector to obtain a fusion anomaly detection result; and taking the abnormal pixel point set with higher confidence in the fusion abnormal detection result as a depth estimation network training sample.
The network training module is used for training the deep estimation network by using the deep estimation network training sample to obtain a trained deep estimation network; inputting the spectral curve into the trained depth estimation network to obtain an object depth estimation result, and performing down-sampling according to the object depth estimation result to obtain a classification training sample; training the classification network by using the classification training sample to obtain a trained classification network; inputting the spectral curve into the trained classification network to obtain a classification result; taking all target pixel points in the classification result as the depth estimation network training set, and performing next round of training on the depth estimation network and the classification network until a preset condition is reached to obtain an underwater object detection model; the preset condition is that parameters of the depth estimation network and the classification network are converged or reach a preset training round number.
And the underwater object detection module to be detected is used for acquiring the hyperspectral water body image to be detected, processing the hyperspectral water body image to be detected to obtain a spectral curve to be detected, and inputting the spectral curve to be detected into the underwater object detection model to obtain an underwater object detection result.
According to the underwater object detection method, the device, the computer equipment and the storage medium, the underwater object detection network is constructed by acquiring the hyperspectral water body image, the underwater object detection network comprises the joint anomaly detector, the depth estimation network and the classification network, the joint anomaly detector is utilized to extract target pixel points with high confidence level from the input hyperspectral water body image, and the influence of background pixels on a detection result is inhibited while a separation result with high robustness is obtained; and then training the depth estimation network by using the extracted abnormal pixel point set to obtain depth prediction information, constructing a classification network training set by using the prediction result of the depth estimation network and a down-sampling method, training the classification network by using the classification network training set to obtain a trained classification network, processing the input image by using the trained classification network, and classifying the result. The method can detect and identify the specific underwater object in any scene.
Drawings
FIG. 1 is a schematic flow chart of a method for detecting underwater objects in one embodiment;
FIG. 2 is a schematic view of an underwater object detection process in another embodiment;
FIG. 3 is a detailed view of the experimental data set in another embodiment;
FIG. 4 is a comparison graph of the detection result and the actual value of the object position according to another embodiment of the present invention;
FIG. 5 is a block diagram showing the structure of an underwater object detection device according to an embodiment;
FIG. 6 is a diagram illustrating an internal structure of a computer device according to an embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
In one embodiment, as shown in fig. 1, there is provided an underwater object detection method comprising the steps of:
step 100: and acquiring a hyperspectral water body image, and processing the hyperspectral water body image to obtain a spectral curve.
The hyperspectral water body image is processed by utilizing the existing water body inherent optical property inversion algorithm.
Step 102: and constructing an underwater object detection network, wherein the underwater object detection network comprises a joint anomaly detector, a depth estimation network and a classification network.
The underwater object detection network is used for processing the input hyperspectral water body image to obtain an underwater object detection result.
And the joint anomaly detector is used for processing the input image through different anomaly detection modules to find out the anomaly pixels in the input image, and performing decision fusion on all anomaly detection results by using an integrated learning algorithm to obtain a fused anomaly detection result.
And the depth estimation network is used for receiving the fusion anomaly detection result and processing the fusion anomaly detection result to obtain depth estimation information.
The classification network is used for receiving a data set obtained by adopting a down-sampling method on the prediction result of the depth estimation network and outputting a final underwater object detection result.
Step 104: and inputting the spectrum curve into a joint anomaly detector to obtain a fusion anomaly detection result.
The joint anomaly detector is used for processing the input hyperspectral image by adopting an anomaly detection module constructed by a plurality of different anomaly detection algorithms and ensuring that anomaly points in the image are extracted in different ways from different angles. And then, fusing different abnormal detection results by means of a decision fusion related research theory in ensemble learning to obtain a fusion abnormal detection result, and ensuring that the fusion abnormal detection result has better robustness.
Step 106: and taking the abnormal pixel point set with higher confidence in the fusion abnormal detection result as a depth estimation network training sample.
Step 108: and training the deep estimation network by using the deep estimation network training sample to obtain the trained deep estimation network.
Inputting a depth estimation training sample into a depth estimation network, extracting spectral characteristics of an input image, realizing estimation of depth information, inputting a depth information estimation result into a water body measurement model to calculate a reconstruction error by a spectral curve reconstruction ratio, optimizing network parameters by taking the reconstruction error as a target function and adding a depth information amplitude constraint, and performing network training in an unsupervised mode, thereby improving the estimation capability of the depth information of the object by the method.
Step 110: and inputting the spectral curve into a trained depth estimation network to obtain an object depth estimation result, and performing down-sampling according to the object depth estimation result to obtain a classification training sample.
Step 112: and training the classification network by using the classification training sample to obtain the trained classification network.
Step 114: inputting the spectrum curve into a trained classification network to obtain a classification result; all target pixel points in the classification result are used as a depth estimation network training set, and the depth estimation network is trained in the next round until a preset condition is reached, so that an underwater object detection model is obtained; the preset condition is that the parameters of the depth estimation network and the classification network are converged or reach the preset training round number.
Step 116: and acquiring a hyperspectral water body image to be detected, processing the hyperspectral water body image to obtain a spectral curve to be detected, and inputting the spectral curve to be detected into the underwater object detection model to obtain an underwater object detection result.
Considering the fitting problem of the depth estimation network and the classification network caused by the sample imperfection problem, the training of the whole network can be completed in an iterative mode by utilizing the output interaction relation of the two networks, so that the object depth estimation and the object detection are combined.
In the method for detecting the underwater object, an underwater object detection network is constructed by acquiring a hyperspectral water body image, the underwater object detection network comprises a joint anomaly detector, a depth estimation network and a classification network, target pixel points with high confidence level are extracted from the input hyperspectral water body image by using the joint anomaly detector, and the influence of background pixels on a detection result is also inhibited while a strong robustness separation result is acquired; and then training the depth estimation network by using the extracted abnormal pixel point set to obtain depth prediction information, constructing a classification network training set by using the prediction result of the depth estimation network and a down-sampling method, training the classification network by using the classification network training set to obtain a trained classification network, processing the input image by using the trained classification network, and classifying the result. The method can detect and identify the specific underwater object in any scene.
In one embodiment, the joint anomaly detector comprises a plurality of different anomaly detection modules and a decision fusion module; various anomaly detection modules include: RX abnormity detection module, LRX abnormity detection module, CRD abnormity detection module and CBAD abnormity detection module. Step 104 further comprises: inputting the spectrum curve into an RX anomaly detection module to obtain an RX anomaly detection result; inputting the spectral curve into an LRX anomaly detection module to obtain an LRX anomaly detection result; inputting the spectrum curve into a CRD abnormity detection module to obtain a CRD abnormity detection result; inputting the spectrum curve into a CBAD anomaly detection module to obtain a CBAD anomaly detection result; and inputting the RX abnormal detection result, the LRX abnormal detection result, the CRD abnormal detection result and the CBAD abnormal detection result into a decision fusion module, and performing decision fusion by adopting an integrated learning algorithm to obtain a fusion abnormal detection result.
In one embodiment, the depth estimation network comprises a first convolutional neural network, a fully-connected network and a water body model; step 108 further comprises: inputting a depth estimation network training sample into a first convolution neural network to obtain a nonlinear convolution spectral feature; inputting the nonlinear convolution spectral characteristics into a full-connection network to obtain object depth estimation; inputting the depth estimation of the object into a water model to obtain a reconstructed spectrum curve; and taking the difference value between the reconstructed spectral curve and the input spectral curve as a target function, adding object depth information numerical value constraint, and performing reverse training on the depth estimation network by utilizing a depth network training and gradient descent algorithm until the weight parameter of the depth estimation network is converged or reaches a preset training round number to obtain the trained depth estimation network.
The first convolutional neural network is a one-dimensional convolutional neural network.
And constructing a depth estimation network through a one-dimensional convolutional neural network, a fully-connected network and a water model.
And estimating the depth of the object according to the spectral characteristics of the input pixel points.
Extracting the spectral characteristics of an input spectral curve through a one-dimensional convolutional neural network, unfolding the acquired spectral characteristics into one-dimensional spectral vectors, and inputting the one-dimensional spectral vectors into a full-connection layer network to realize depth information estimation; and bringing the estimated depth into a water body measurement model for spectrum curve reconstruction.
Since the optical properties related to the water body have been solved in step 100, and the fingerprint characteristics of the object to be detected are known a priori, the input spectral curve can be reconstructed as long as the estimated depth information is input into the water body measurement model.
Calculating a difference value between a reconstructed spectral curve and an input spectral curve, taking the difference value as a target function, adding object depth information numerical value constraint, and adjusting weight parameters of a depth estimation network by utilizing a depth estimation network training set and a gradient descent algorithm; when the value of the depth information exceeds a certain value, the change of the depth information does not affect the spectral characteristics of the target in the water body any more, and at the moment, the gradient of the target function is 0, and the network parameters cannot be adjusted through a gradient descent algorithm. Therefore, in order to ensure that the network parameters can be converged well, a corresponding depth information amplitude constraint is added to the objective function.
In one embodiment, the first convolutional neural network comprises a one-dimensional convolutional layer, a batch normalization layer and a pooling layer; the fully connected network includes a flatten layer and a fully connected layer. Step 108 further comprises: inputting a depth estimation network training sample into a first convolutional neural network, and extracting the features of the depth estimation network training sample by using a one-dimensional convolutional layer with the receptive field size of 1 multiplied by 3, the step length of 1 and the zero padding number of 1 to obtain convolutional features; carrying out batch normalization processing on the convolution characteristics to obtain normalized convolution characteristics; and utilizing the pooling layer to carry out down-sampling on the normalized convolution characteristics to obtain nonlinear convolution spectral characteristics.
In one embodiment, the fully connected network includes a flatten layer and a fully connected layer. Step 108 further comprises: inputting the nonlinear convolution spectral features into a flatten layer to obtain flattened spectral features; inputting the flattened spectral characteristics into a full-connection layer for transformation to obtain object depth estimation.
In one embodiment, step 110 further comprises: inputting the spectral curve into a trained depth estimation network to obtain the corresponding depth of each pixel point of the hyperspectral water body image, and classifying the image according to depth information to obtain a target pixel set and a background pixel set; and obtaining the same number of target pixel points and background pixel points by utilizing a downsampling mode and constructing a classification training sample according to the target pixel points and the background pixel points.
Predicting the depth corresponding to each pixel point of the input image by using the trained depth estimation network, and dividing the image pixels into two target and background categories according to the depth information; solving the whole input hyperspectral image through a depth estimation network, wherein the depth information of target pixel points is far smaller than that corresponding to background pixel points, and the number of the target pixel points is far greater than that of the background pixel points; and acquiring the same number of target pixel points and background pixel points by utilizing a downsampling mode and constructing a training data set according to the target pixel points and the background pixel points.
Because the target pixel points usually only occupy a small part of pixels in the whole image, the number of the target pixel points is far smaller than that of the background samples, and finally the problem of class imbalance is caused, so that a classification training set with the same number of target samples and background samples needs to be constructed in a downsampling mode.
In one embodiment, step 112 further comprises: inputting the classified training samples into a second convolutional neural network to obtain nonlinear convolutional spectral characteristics; inputting the nonlinear convolution spectral characteristics into a full-connection network to obtain pixel category estimation; and carrying out reverse training on the classification network by adopting a gradient descent algorithm according to the pixel class estimation and classification training samples until the classification network is converged to obtain the trained classification network.
An anomaly detector of an underwater object detection network extracts abnormal pixel features of an input hyperspectral image, carries out decision fusion, uses an obtained fusion anomaly detection result as a depth estimation network training sample, trains a depth estimation network by using the depth estimation network to obtain a trained depth estimation network, processes a hyperspectral water body image by using the trained depth estimation network, determines a classification training sample, trains the classification network by using the classification training sample to obtain a trained classification network, processes the whole hyperspectral water body image by using the trained classification network, determines a training sample of the depth estimation network, carries out next round of training on the depth estimation network by using the training sample of the depth estimation network until a preset condition is reached to obtain an underwater object detection model, and realizes detection on the hyperspectral water body image by using the underwater object detection model, and obtaining the detection result of the underwater object.
In another embodiment, a process for detecting an underwater object is provided, which is shown in fig. 2 and includes the following steps:
the first step is as follows: and (3) taking the hyperspectral water body image subjected to a series of preprocessing operations such as atmospheric correction and the like as the input of the network.
The second step is that: processing the input image by different anomaly detection algorithms to find out the abnormal pixels in the input image, and performing decision fusion on all the anomaly detection results by using an integrated learning algorithm;
2.1, processing the input hyperspectral image through an RX anomaly detection algorithm to obtain a corresponding anomaly detection result;
2.2, processing the input hyperspectral image through an LRX anomaly detection algorithm to obtain a corresponding anomaly detection result;
2.3, processing the input hyperspectral image through a CRD anomaly detection algorithm to obtain a corresponding anomaly detection result;
2.4, processing the input hyperspectral image through a CBAD anomaly detection algorithm to obtain a corresponding anomaly detection result;
and 2.5, performing decision fusion on all the abnormal detection results by using a voting theory in ensemble learning.
The third step: and extracting an abnormal pixel point set with higher confidence in the fusion abnormal detection result as a depth estimation network training data set.
The fourth step: an encoder in a depth estimation network is constructed through a one-dimensional convolution neural network and a full-connection network, and object depth estimation is carried out according to spectral characteristics of input pixel points
4.1, extracting the characteristics of the original hyperspectral curve by using a one-dimensional convolution layer with the receptive field size of 1 multiplied by 3, the step length of 1 and the zero padding number size of 1;
4.2, performing batch normalization operation on the features extracted from the one-dimensional convolutional layer;
4.3, utilizing the pooling layer to perform down-sampling operation with the sampling step length of 2 on the features after batch processing;
4.4 repeating the step 4.1-4.3 three times, and extracting nonlinear convolution spectral characteristics;
4.5 flattening the extracted convolution spectral characteristics by using a flatten layer;
and 4.6, transforming the flattened spectral features through a full-connection network to realize the estimation of the depth information.
The fifth step: bringing the estimated depth into a water body measurement model for spectral curve reconstruction;
and a sixth step: calculating a difference value between the reconstructed spectrum curve and the input spectrum curve, taking the difference value as a target function, adding object depth information numerical value constraint, and adjusting weight parameters of the depth prediction sub-network in the fourth step by utilizing a depth estimation network training set and a gradient descent algorithm;
the seventh step: repeating the steps 4-6 until the network parameters converge or reach a preset training round number;
eighth step: predicting the depth corresponding to each pixel point of the input image by using the trained depth estimation network, and dividing the image pixels into two target and background categories according to the depth information;
setting a threshold value, setting pixel points with depth information smaller than the threshold value as target pixels, and setting pixel points with depth information larger than the threshold value as background pixel points; the hyper-parameters pertaining to the present invention regarding the setting of the threshold values need to be set according to the specifics of the data set.
The ninth step: and acquiring the same number of target pixel points and background pixel points by utilizing a downsampling mode and constructing a training data set according to the target pixel points and the background pixel points.
The tenth step: and (4) constructing a classifier by using the one-dimensional convolutional neural network and the fully-connected network, and training the classification network by using the training set obtained in the step nine.
10.1, extracting the characteristics of the original hyperspectral curve by using a one-dimensional convolution layer with the receptive field size of 1 multiplied by 3, the step length of 1 and the zero padding number size of 1;
10.2, performing batch normalization operation on the features extracted from the one-dimensional convolutional layer;
10.3 utilizing the pooling layer to perform down-sampling operation with the sampling step length of 2 on the features after batch processing;
10.4 repeating the step for 10.1 to 10.3 times, and extracting nonlinear convolution spectral features;
10.5 flattening the extracted convolution spectral characteristics by using a flatten layer;
10.6, the flattened spectral features are transformed through a full-connection network to realize the estimation of pixel types;
10.7 calculating the classification error between the class prediction result and the label and updating the network parameter weight for the objective function through a gradient descent algorithm;
10.8 repeat steps 10.1-10.7 until the classification network converges.
Step eleven, processing the whole hyperspectral image by using a classification network, forming a depth estimation network training set by all target pixel points in a classification result, inputting the depth estimation network training set into the depth estimation network, and training the depth estimation network;
and a twelfth step of repeating the steps 4-11 until the parameters of the depth estimation network and the classification network converge or reach a preset training round number.
In a verification embodiment, a correlation experiment is carried out on the obtained real water body data set, and the difference between the detection result of the invention and the real value of the object position can reflect the quality of the prediction performance of the invention. Fig. 3 is a specific case of an experimental data set, in which fig. 3(a) is a scene diagram of the data set, fig. 3(b) is a schematic diagram of a target position, and fig. 3(c) is a spectrum graph of a target to be detected. Fig. 4 is a comparison graph between the detection result and the true value of the object position according to the present invention, in which fig. 4(a) is a scene graph of a data set, fig. 4(b) is a schematic diagram of the target position, and fig. 4(c) is the detection result according to the present invention.
It should be understood that although the various steps in the flow charts of fig. 1-2 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least some of the steps in fig. 1-2 may include multiple sub-steps or multiple stages that are not necessarily performed at the same time, but may be performed at different times, and the order of performance of the sub-steps or stages is not necessarily sequential, but may be performed in turn or alternating with other steps or at least some of the sub-steps or stages of other steps.
In one embodiment, as shown in fig. 5, there is provided an underwater object detection apparatus including: image data acquisition module, network construction module, training sample construction module, network training module and the underwater object detection module that awaits measuring, wherein:
the image data acquisition module is used for acquiring a hyperspectral water body image and processing the hyperspectral water body image to obtain a spectral curve.
The network construction module is used for constructing an underwater object detection network, and the underwater object detection network comprises a joint anomaly detector, a depth estimation network and a classification network.
The training sample construction module is used for inputting the spectrum curve into the combined anomaly detector to obtain a fusion anomaly detection result; and taking the abnormal pixel point set with higher confidence in the fusion abnormal detection result as a depth estimation network training sample.
The network training module is used for training the depth estimation network by using the depth estimation network training sample to obtain a trained depth estimation network; inputting the spectral curve into a trained depth estimation network to obtain an object depth estimation result, and performing down-sampling according to the object depth estimation result to obtain a classification training sample; training the classification network by using the classification training sample to obtain a trained classification network; inputting the spectrum curve into a trained classification network to obtain a classification result; all target pixel points in the classification result are used as a depth estimation network training set, and the depth estimation network is trained in the next round until a preset condition is reached, so that an underwater object detection model is obtained; the preset condition is that the parameters of the depth estimation network and the classification network are converged or reach the preset training round number.
And the underwater object detection module to be detected is used for acquiring the hyperspectral water body image to be detected, processing the hyperspectral water body image to be detected to obtain a spectral curve to be detected, and inputting the spectral curve to be detected into the underwater object detection model to obtain an underwater object detection result.
In one embodiment, the joint anomaly detector comprises a plurality of different anomaly detection modules and a decision fusion module; various anomaly detection modules include: RX abnormity detection module, LRX abnormity detection module, CRD abnormity detection module and CBAD abnormity detection module.
The training sample construction module is also used for inputting the spectral water body image into the RX anomaly detection module to obtain an RX anomaly detection result; inputting the spectral curve into an LRX anomaly detection module to obtain an LRX anomaly detection result; inputting the spectrum curve into a CRD abnormity detection module to obtain a CRD abnormity detection result; inputting the spectrum curve into a CBAD anomaly detection module to obtain a CBAD anomaly detection result; and inputting the RX abnormal detection result, the LRX abnormal detection result, the CRD abnormal detection result and the CBAD abnormal detection result into a decision fusion module, and performing decision fusion by adopting an integrated learning algorithm to obtain a fusion abnormal detection result.
In one embodiment, the depth estimation network comprises a first convolutional neural network, a fully-connected network and a water body model; the network training module is also used for inputting the depth estimation network training sample into the first convolution neural network to obtain the nonlinear convolution spectral characteristics; inputting the nonlinear convolution spectral characteristics into a full-connection network to obtain object depth estimation; inputting the depth estimation of the object into a water model to obtain a reconstructed spectrum curve; and taking the difference value between the reconstructed spectral curve and the input spectral curve as a target function, adding object depth information numerical value constraint, and performing reverse training on the depth estimation network by utilizing a depth network training and gradient descent algorithm until the weight parameter of the depth estimation network is converged or reaches a preset training round number to obtain the trained depth estimation network.
In one embodiment, the first convolutional neural network comprises a one-dimensional convolutional layer, a batch normalization layer and a pooling layer; the fully connected network includes a flatten layer and a fully connected layer. The network training module is also used for inputting the depth estimation network training sample into a first convolution neural network, and extracting the characteristics of the depth estimation network training sample by using a one-dimensional convolution layer with the receptive field size of 1 multiplied by 3, the step length of 1 and the zero padding number size of 1 to obtain convolution characteristics; carrying out batch normalization processing on the convolution characteristics to obtain normalized convolution characteristics; and utilizing the pooling layer to carry out down-sampling on the normalized convolution characteristics to obtain nonlinear convolution spectral characteristics.
In one embodiment, the fully connected network includes a flatten layer and a fully connected layer. The network training module is also used for inputting the nonlinear convolution spectral features into a flatten layer to obtain flattened spectral features; inputting the flattened spectral characteristics into a full-connection layer for transformation to obtain object depth estimation.
In one embodiment, the network training module is further configured to input the spectral curve into a trained depth estimation network to obtain a depth corresponding to each pixel point of the hyperspectral water body image, and classify the image according to depth information to obtain a target pixel set and a background pixel set; and obtaining the same number of target pixel points and background pixel points by utilizing a downsampling mode and constructing a classification training sample according to the target pixel points and the background pixel points.
In one embodiment, the network training module is further configured to input the classification training samples into a second convolutional neural network to obtain nonlinear convolutional spectral features; inputting the nonlinear convolution spectral characteristics into a full-connection network to obtain pixel category estimation; and carrying out reverse training on the classification network by adopting a gradient descent algorithm according to the pixel class estimation and classification training samples until the classification network is converged to obtain the trained classification network.
For specific definition of the underwater object detection device, reference may be made to the above definition of the underwater object detection method, which is not described herein again. All or part of the modules in the underwater object detection device can be realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.
In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as shown in fig. 6. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a method of underwater object detection. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell of the computer equipment, an external keyboard, a touch pad or a mouse and the like.
Those skilled in the art will appreciate that the architecture shown in fig. 6 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In an embodiment, a computer device is provided, comprising a memory storing a computer program and a processor implementing the steps of the method in the above embodiments when the processor executes the computer program.
In an embodiment, a computer-readable storage medium is provided, on which a computer program is stored, which computer program, when being executed by a processor, carries out the steps of the method in the above-mentioned embodiments.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (10)
1. An underwater object detection method, characterized in that the method comprises:
acquiring a hyperspectral water body image, and processing the hyperspectral water body image to obtain a spectral curve;
constructing an underwater object detection network, wherein the underwater object detection network comprises a joint anomaly detector, a depth estimation network and a classification network;
inputting the spectrum curve into a joint anomaly detector to obtain a fusion anomaly detection result;
taking an abnormal pixel point set with higher confidence in the fusion abnormal detection result as a depth estimation network training sample;
training the deep estimation network by using the deep estimation network training sample to obtain a trained deep estimation network;
inputting the spectral curve into the trained depth estimation network to obtain an object depth estimation result, and performing down-sampling according to the object depth estimation result to obtain a classification training sample;
training the classification network by using the classification training sample to obtain a trained classification network;
inputting the spectral curve into the trained classification network to obtain a classification result; taking all target pixel points in the classification result as the depth estimation network training set, and performing next round of training on the depth estimation network and the classification network until a preset condition is reached to obtain an underwater object detection model; the preset condition is that the parameters of the depth estimation network and the classification network are converged or reach a preset training round number;
and acquiring a hyperspectral water body image to be detected, processing the hyperspectral water body image to obtain a spectral curve to be detected, and inputting the spectral curve to be detected into the underwater object detection model to obtain an underwater object detection result.
2. The method of claim 1, wherein the joint anomaly detector comprises a plurality of different anomaly detection modules and a decision fusion module; the plurality of different anomaly detection modules includes: the system comprises an RX abnormity detection module, an LRX abnormity detection module, a CRD abnormity detection module and a CBAD abnormity detection module;
inputting the spectrum curve into a joint anomaly detector to obtain a fusion anomaly detection result; the method comprises the following steps:
inputting the spectrum curve into the RX abnormity detection module to obtain an RX abnormity detection result;
inputting the spectral curve into the LRX anomaly detection module to obtain an LRX anomaly detection result;
inputting the spectrum curve into the CRD abnormity detection module to obtain a CRD abnormity detection result;
inputting the spectrum curve into the CBAD abnormity detection module to obtain a CBAD abnormity detection result;
and inputting the RX abnormity detection result, the LRX abnormity detection result, the CRD abnormity detection result and the CBAD abnormity detection result into the decision fusion module, and performing decision fusion by adopting an ensemble learning algorithm to obtain a fusion abnormity detection result.
3. The method of claim 1, wherein the depth estimation network comprises a first convolutional neural network, a fully-connected network, and a water body model;
training the deep estimation network by using the deep estimation network training sample to obtain a trained deep estimation network, wherein the training comprises the following steps:
inputting the depth estimation network training sample into the first convolution neural network to obtain a nonlinear convolution spectral feature;
inputting the nonlinear convolution spectral features into the full-connection network to obtain object depth estimation;
inputting the object depth estimation into the water body model to obtain a reconstructed spectrum curve;
and taking the difference value between the reconstructed spectral curve and the input spectral curve as a target function, adding object depth information numerical value constraint, and performing reverse training on the depth estimation network by utilizing a depth network training and gradient descent algorithm until the weight parameter of the depth estimation network is converged or reaches a preset training round number to obtain the trained depth estimation network.
4. The method of claim 3, wherein the first convolutional neural network comprises a one-dimensional convolutional layer, a batch normalization layer, and a pooling layer; the fully connected network comprises a flatten layer and a fully connected layer;
inputting the depth estimation network training sample into the first convolutional neural network to obtain a nonlinear convolutional spectrum feature, including:
inputting the depth estimation network training sample into a first convolutional neural network, and performing feature extraction on the depth estimation network training sample by using a one-dimensional convolutional layer with the receptive field size of 1 multiplied by 3, the step length of 1 and the zero padding number size of 1 to obtain convolutional features;
carrying out batch normalization processing on the convolution characteristics to obtain normalized convolution characteristics;
and utilizing the pooling layer to carry out down-sampling on the normalized convolution characteristics to obtain nonlinear convolution spectral characteristics.
5. The method of claim 3, wherein the fully connected network comprises a flatten layer and a fully connected layer;
inputting the nonlinear convolution spectral features into the fully-connected network to obtain an object depth estimate, comprising:
inputting the nonlinear convolution spectral features into a flatten layer to obtain flattened spectral features;
and inputting the flattened spectral features into the full-connection layer for transformation to obtain object depth estimation.
6. The method according to claim 1, wherein inputting the spectral curve into the trained depth estimation network to obtain an object depth estimation result, and obtaining a classification training sample by using a down-sampling method according to the object depth estimation result, comprises:
inputting the spectral curve into the trained depth estimation network to obtain the corresponding depth of each pixel point of the hyperspectral water body image, and classifying the image according to depth information to obtain a target pixel set and a background pixel set;
and obtaining the same number of target pixel points and background pixel points by utilizing a downsampling mode and constructing a classification training sample according to the target pixel points and the background pixel points.
7. The method of claim 1, wherein training the classification network with the classification training samples to obtain a trained classification network comprises:
inputting the classified training samples into a second convolutional neural network to obtain nonlinear convolutional spectral characteristics;
inputting the nonlinear convolution spectral features into the full-connection network to obtain pixel category estimation;
and carrying out reverse training on the classification network by adopting a gradient descent algorithm according to the pixel class estimation and the classification training sample until the classification network is converged to obtain the trained classification network.
8. An underwater object detection device, the device comprising:
the image data acquisition module is used for acquiring a hyperspectral water body image and processing the hyperspectral water body image to obtain a spectral curve;
the underwater object detection network comprises a joint anomaly detector, a depth estimation network and a classification network;
the training sample construction module is used for inputting the spectral curve into a joint anomaly detector to obtain a fusion anomaly detection result; taking an abnormal pixel point set with higher confidence in the fusion abnormal detection result as a depth estimation network training sample;
the network training module is used for training the deep estimation network by using the deep estimation network training sample to obtain a trained deep estimation network; inputting the spectral curve into the trained depth estimation network to obtain an object depth estimation result, and performing down-sampling according to the object depth estimation result to obtain a classification training sample; training the classification network by using the classification training sample to obtain a trained classification network; inputting the spectral curve into the trained classification network to obtain a classification result; taking all target pixel points in the classification result as the depth estimation network training set, and performing next round of training on the depth estimation network and the classification network until a preset condition is reached to obtain an underwater object detection model; the preset condition is that the parameters of the depth estimation network and the classification network are converged or reach a preset training round number;
and acquiring a hyperspectral water body image to be detected, processing the hyperspectral water body image to obtain a spectral curve to be detected, and inputting the spectral curve to be detected into the underwater object detection model to obtain an underwater object detection result.
9. A computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor implements the steps of the method of any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110418252.6A CN112990106B (en) | 2021-04-19 | 2021-04-19 | Underwater object detection method, device, computer equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110418252.6A CN112990106B (en) | 2021-04-19 | 2021-04-19 | Underwater object detection method, device, computer equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112990106A true CN112990106A (en) | 2021-06-18 |
| CN112990106B CN112990106B (en) | 2022-09-09 |
Family
ID=76341097
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110418252.6A Active CN112990106B (en) | 2021-04-19 | 2021-04-19 | Underwater object detection method, device, computer equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112990106B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113392795A (en) * | 2021-06-29 | 2021-09-14 | 北京百度网讯科技有限公司 | Joint detection model training method, joint detection device, joint detection equipment and joint detection medium |
| CN117132473A (en) * | 2023-10-20 | 2023-11-28 | 中国海洋大学 | Underwater rare earth spectrum detection method and spectrum super-resolution reconstruction model building method thereof |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108765460A (en) * | 2018-05-22 | 2018-11-06 | 北京航空航天大学 | Space-time joint method for detecting abnormality based on high spectrum image and electronic equipment |
| CN109871830A (en) * | 2019-03-15 | 2019-06-11 | 中国人民解放军国防科技大学 | Spatial-spectral fusion hyperspectral image classification method based on three-dimensional depth residual error network |
| CN110852227A (en) * | 2019-11-04 | 2020-02-28 | 中国科学院遥感与数字地球研究所 | Hyperspectral image deep learning classification method, device, equipment and storage medium |
| CN111695469A (en) * | 2020-06-01 | 2020-09-22 | 西安电子科技大学 | Hyperspectral image classification method of lightweight depth separable convolution feature fusion network |
| CN112308152A (en) * | 2020-11-03 | 2021-02-02 | 西安电子科技大学 | Hyperspectral image ground object classification method based on spectrum segmentation and homogeneous region detection |
-
2021
- 2021-04-19 CN CN202110418252.6A patent/CN112990106B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108765460A (en) * | 2018-05-22 | 2018-11-06 | 北京航空航天大学 | Space-time joint method for detecting abnormality based on high spectrum image and electronic equipment |
| CN109871830A (en) * | 2019-03-15 | 2019-06-11 | 中国人民解放军国防科技大学 | Spatial-spectral fusion hyperspectral image classification method based on three-dimensional depth residual error network |
| CN110852227A (en) * | 2019-11-04 | 2020-02-28 | 中国科学院遥感与数字地球研究所 | Hyperspectral image deep learning classification method, device, equipment and storage medium |
| CN111695469A (en) * | 2020-06-01 | 2020-09-22 | 西安电子科技大学 | Hyperspectral image classification method of lightweight depth separable convolution feature fusion network |
| CN112308152A (en) * | 2020-11-03 | 2021-02-02 | 西安电子科技大学 | Hyperspectral image ground object classification method based on spectrum segmentation and homogeneous region detection |
Non-Patent Citations (2)
| Title |
|---|
| 刘启超等: "SSCDenseNet:一种空-谱卷积稠密网络的高光谱图像分类算法", 《电子学报》 * |
| 郑庆翔等: "基于深度卷积神经网络的高光谱图像分类算法研究", 《白城师范学院学报》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113392795A (en) * | 2021-06-29 | 2021-09-14 | 北京百度网讯科技有限公司 | Joint detection model training method, joint detection device, joint detection equipment and joint detection medium |
| CN113392795B (en) * | 2021-06-29 | 2023-07-04 | 北京百度网讯科技有限公司 | Combined detection model training, combined detection method, device, equipment and medium |
| CN117132473A (en) * | 2023-10-20 | 2023-11-28 | 中国海洋大学 | Underwater rare earth spectrum detection method and spectrum super-resolution reconstruction model building method thereof |
| CN117132473B (en) * | 2023-10-20 | 2024-01-23 | 中国海洋大学 | Underwater rare earth spectrum detection method and spectrum super-resolution reconstruction model building method thereof |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112990106B (en) | 2022-09-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Makantasis et al. | Tensor-based classification models for hyperspectral data analysis | |
| CN110941986B (en) | Living body detection model training method, living body detection model training device, computer equipment and storage medium | |
| Wadoux et al. | Multi-source data integration for soil mapping using deep learning | |
| CN108846355B (en) | Image processing method, face recognition device and computer equipment | |
| US11094040B2 (en) | Noise detection method for time-series vegetation index derived from remote sensing images | |
| Zhang et al. | Feature-level change detection using deep representation and feature change analysis for multispectral imagery | |
| US11429837B2 (en) | Sparse neural network based anomaly detection in multi-dimensional time series | |
| Mercier et al. | Conditional copulas for change detection in heterogeneous remote sensing images | |
| Moser et al. | Unsupervised change detection from multichannel SAR data by Markovian data fusion | |
| CN111950329A (en) | Target detection and model training method and device, computer equipment and storage medium | |
| Karuppusamy | Building detection using two-layered novel convolutional neural networks | |
| CN112990106B (en) | Underwater object detection method, device, computer equipment and storage medium | |
| CN111199523B (en) | Power equipment identification method, device, computer equipment and storage medium | |
| CN113392931B (en) | Hyperspectral open set classification method based on self-supervision learning and multitask learning | |
| CN110956615B (en) | Image quality evaluation model training method and device, electronic equipment and storage medium | |
| Huyck et al. | Towards rapid citywide damage mapping using neighborhood edge dissimilarities in very high-resolution optical satellite imagery—Application to the 2003 Bam, Iran, earthquake | |
| CN111967535B (en) | Fault diagnosis method and device for temperature sensor of grain storage management scene | |
| Haque et al. | Spectral-spatial feature extraction using PCA and multi-scale deep convolutional neural network for hyperspectral image classification | |
| CN111008643A (en) | Image classification method and device based on semi-supervised learning and computer equipment | |
| CN112990107B (en) | Hyperspectral remote sensing image underwater target detection method and device and computer equipment | |
| Vayadande et al. | Cyclone Intensity Estimation on INSAT 3D IR Imagery Using Deep Learning | |
| CN111105417B (en) | Image noise positioning method and system | |
| CN113313169A (en) | Training material intelligent identification method, device and equipment based on deep learning | |
| Ebert-Uphoff et al. | Evaluation, tuning and interpretation of neural networks for meteorological applications | |
| Rezvanian et al. | Patch-based sparse and convolutional autoencoders for anomaly detection in hyperspectral images |
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 |