CN117611983B - Underwater target detection method and system based on hidden communication technology and deep learning - Google Patents
Underwater target detection method and system based on hidden communication technology and deep learning Download PDFInfo
- Publication number
- CN117611983B CN117611983B CN202311537443.XA CN202311537443A CN117611983B CN 117611983 B CN117611983 B CN 117611983B CN 202311537443 A CN202311537443 A CN 202311537443A CN 117611983 B CN117611983 B CN 117611983B
- Authority
- CN
- China
- Prior art keywords
- information
- image
- target
- secret
- secret information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000004891 communication Methods 0.000 title claims abstract description 81
- 238000001514 detection method Methods 0.000 title claims abstract description 59
- 238000005516 engineering process Methods 0.000 title claims abstract description 31
- 238000013135 deep learning Methods 0.000 title claims abstract description 28
- 238000000034 method Methods 0.000 claims abstract description 53
- 238000012545 processing Methods 0.000 claims abstract description 35
- 230000005540 biological transmission Effects 0.000 claims abstract description 28
- 238000007781 pre-processing Methods 0.000 claims abstract description 22
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 13
- 238000000605 extraction Methods 0.000 claims description 30
- 239000011159 matrix material Substances 0.000 claims description 27
- 238000013139 quantization Methods 0.000 claims description 16
- 238000012546 transfer Methods 0.000 claims description 5
- 230000001502 supplementing effect Effects 0.000 claims description 4
- 238000012856 packing Methods 0.000 claims description 3
- 230000001360 synchronised effect Effects 0.000 claims 1
- 238000012549 training Methods 0.000 description 28
- 230000008569 process Effects 0.000 description 19
- 230000006870 function Effects 0.000 description 9
- 238000002372 labelling Methods 0.000 description 6
- 238000012795 verification Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 238000011068 loading method Methods 0.000 description 4
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 230000004927 fusion Effects 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 3
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000013480 data collection Methods 0.000 description 2
- 230000003247 decreasing effect Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000011049 filling Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000004091 panning Methods 0.000 description 2
- 238000009827 uniform distribution Methods 0.000 description 2
- 238000010521 absorption reaction Methods 0.000 description 1
- 230000004071 biological effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 239000013535 sea water Substances 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L1/00—Arrangements for detecting or preventing errors in the information received
- H04L1/004—Arrangements for detecting or preventing errors in the information received by using forward error control
- H04L1/0056—Systems characterized by the type of code used
- H04L1/0059—Convolutional codes
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/10—Image acquisition
- G06V10/12—Details of acquisition arrangements; Constructional details thereof
- G06V10/14—Optical characteristics of the device performing the acquisition or on the illumination arrangements
- G06V10/141—Control of illumination
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
- G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
- G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
- G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/05—Underwater scenes
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B11/00—Transmission systems employing sonic, ultrasonic or infrasonic waves
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L1/00—Arrangements for detecting or preventing errors in the information received
- H04L1/004—Arrangements for detecting or preventing errors in the information received by using forward error control
- H04L1/0041—Arrangements at the transmitter end
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L1/00—Arrangements for detecting or preventing errors in the information received
- H04L1/004—Arrangements for detecting or preventing errors in the information received by using forward error control
- H04L1/0045—Arrangements at the receiver end
- H04L1/0054—Maximum-likelihood or sequential decoding, e.g. Viterbi, Fano, ZJ algorithms
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L1/00—Arrangements for detecting or preventing errors in the information received
- H04L1/004—Arrangements for detecting or preventing errors in the information received by using forward error control
- H04L1/0056—Systems characterized by the type of code used
- H04L1/0071—Use of interleaving
-
- 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
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Multimedia (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Biodiversity & Conservation Biology (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an underwater target detection method and system based on a hidden communication technology and deep learning, and the method comprises the following steps: acquiring an image of a target object in real time; preprocessing the image, and extracting features of the preprocessed image by using a convolutional neural network model to obtain boundary frame coordinates, category labels and confidence scores of each target; the information transmission method based on the hidden communication realizes the safe transmission of the boundary frame coordinates, the category labels, the confidence scores and the communication information of each target. The invention can effectively solve the problems of real-time image processing and underwater hidden communication of the underwater robot, and is more in line with the complex and changeable environment at present.
Description
Technical Field
The invention belongs to the field of underwater robots, in particular to the field of underwater target detection and hidden communication, and particularly relates to an underwater target detection method and system based on a hidden communication technology and deep learning.
Background
With the development of technology, underwater target detection has become a hot spot technical problem, and due to the specificity of an underwater environment, the acquisition difficulty of an underwater image is much greater than that of a land image, and the acquisition of the underwater image is affected by various factors, such as an underwater complex illumination environment, water quality conditions, similarity of a target and a background, and the like, which are main factors affecting the accuracy of underwater target detection. In recent years, deep learning is being vigorously developed. The object detection technology has greatly progressed and is widely applied to various scenes. However, the network structure of the method in the present stage is complex and the parameter quantity is huge, which is unfavorable for real-time detection.
Underwater communication technology refers to technology for transmitting information in an underwater environment, and includes various forms of acoustic communication, electromagnetic wave communication, optical communication, and the like. In seawater, the absorption and attenuation of electromagnetic waves and light waves propagating in water are serious, and sound waves are currently the main mode of information propagation in water. In an underwater environment, the traditional communication modes such as electromagnetic waves, light waves and the like have many limitations and disadvantages, and cannot meet the actual application demands. The underwater sound hidden communication technology can effectively improve confidentiality and concealment of underwater communication by using means of special modulation and demodulation methods, encryption algorithms and the like, has stronger anti-interference capability, can cope with the influence factors such as ocean background noise, underwater biological activity interference, hostile interference and the like, and ensures accurate transmission of information.
The coding layer hidden communication technology transmits the secret information by embedding the secret information into the coded public information, so as to realize hidden communication. The embedding of secret information affects the error correction performance of channel coding, thus increasing the bit error rate of information transmitted in the original communication system at the receiving end and the risk of the embedded information being detected. In order to reduce the negative influence of secret information embedding on the original communication system, the binary Hamming matrix based embedding method is introduced into the hidden communication of the coding layer, and the embedding method can effectively reduce the influence of secret information on carrier information and improve the concealment and safety of the hidden communication. But the introduction of the binary hamming matrix embedding method increases the bit error rate of the secret information.
Therefore, the invention provides an underwater robot scheme combining a technology for detecting an underwater target in real time and a coding layer hidden communication technology based on binary Hamming matrix embedding improvement for underwater acoustic communication, which can well solve the problem.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides the underwater target detection method and the underwater target detection system based on the hidden communication technology and the deep learning, which can effectively solve the problems of real-time image processing of the underwater robot and underwater hidden communication and are more in line with the complex and changeable environment at present.
In order to achieve the above object, the present invention provides the following solutions:
The underwater target detection method based on the hidden communication technology and the deep learning comprises the following steps:
acquiring an image of a target object in real time;
Preprocessing the image, and extracting features of the preprocessed image by using a convolutional neural network model to obtain boundary frame coordinates, category labels and confidence scores of each target;
The information transmission method based on the hidden communication realizes the safe transmission of the boundary frame coordinates, the category labels, the confidence scores and the communication information of each target.
Preferably, the method for acquiring the image of the target object in real time comprises the following steps:
And acquiring an image of the region to be detected by using the high-definition camera module to obtain an image of the target object.
Preferably, the method for preprocessing the image and extracting features of the preprocessed image by using a convolutional neural network model to obtain the boundary frame coordinates, class labels and confidence scores of each target comprises the following steps:
Adjusting the size of the image of the target object and normalizing the pixel value;
based on YOLOv model, introducing a plurality of fine-grained feature layers on the basis of Darknet, extracting and fusing the feature information of different scales of the preprocessed image, and predicting target frames of different scales by using a plurality of detection heads to obtain the boundary frame coordinates, class labels and confidence scores of each target.
Preferably, the information transmission method based on the hidden communication, the method for realizing the safe transfer of the boundary frame coordinates, the category labels and the confidence scores of each target and the communication information, comprises the following steps:
Performing convolutional encoding operation on the public information to form carrier information; carrying out convolutional encoding operation on the secret information, and carrying out interleaving treatment on the secret information through an interleaver; embedding secret information into public information by using a matrix embedding mode based on binary Hamming codes to obtain secret carrying information; modulating the secret information to obtain modulated signals and transmitting the modulated signals through sonar; wherein, the secret information is information processed by the image processing module;
And carrying out matrix quantization extraction on the secret information, de-interleaving the result of matrix quantization extraction, and then using the result as the input of a Viterbi soft decision decoder to decode and recover the secret information.
The invention also discloses an underwater target detection system based on the hidden communication technology and the deep learning, which comprises: the device comprises an image processing module, an image acquisition module and a hidden communication module;
the image acquisition module is used for acquiring an image of a target object in real time;
The image processing module is used for preprocessing the image, and extracting features of the preprocessed image by using a convolutional neural network model to obtain the boundary frame coordinates, the category labels and the confidence scores of each target;
The hidden communication module is used for realizing the safe transfer of the boundary frame coordinates, the category labels, the confidence scores and the communication information of each target based on the information transmission method of the hidden communication.
Preferably, the image acquisition module includes: the high-definition camera module and the LED light supplementing device;
The high-definition camera module is used for collecting images of the region to be detected and obtaining images of the target object;
The LED light supplementing device is used for assisting the high-definition camera module.
Preferably, the image processing module includes: a preprocessing unit and a feature extraction unit;
The preprocessing unit is used for adjusting the size of the image of the target object and normalizing the pixel value;
The feature extraction unit is used for introducing a plurality of fine-grained feature layers on the basis of a YOLOv model and Darknet, extracting and fusing feature information of different scales of the preprocessed image, and predicting target frames of different scales by using a plurality of detection heads to obtain the boundary frame coordinates, category labels and confidence scores of each target.
Preferably, the covert communication module includes: a transmitter, a receiver, a sonar, a signal processing unit and a control unit;
The signal processing unit is used for performing convolution encoding operation on the public information to form carrier information; carrying out convolutional encoding operation on the secret information, and carrying out interleaving treatment on the secret information through an interleaver; embedding secret information into public information by using a matrix embedding mode based on binary Hamming codes to obtain secret carrying information; modulating the secret information to obtain a modulated signal; wherein, the secret information is information processed by the image processing module;
the transmitter is used for transmitting the secret carrying information;
The sonar is used for sending out the modulated signals;
The receiver is used for receiving the data signal from the transmitter;
The control unit is used for carrying out matrix quantization extraction on the secret information, de-interleaving the result of matrix quantization extraction, and then taking the result as the input of the Viterbi soft decision decoder, and decoding and recovering the secret information.
The invention also discloses an underwater target detection robot based on the hidden communication technology and the deep learning, and the robot applies the underwater target detection method based on the hidden communication technology and the deep learning.
Compared with the prior art, the invention has the beneficial effects that:
The image processing module disclosed by the invention is based on YOLOv algorithm, adopts DarkNet53 as a feature extractor, introduces a multi-scale prediction and feature fusion strategy, can better capture target information of different scales, can effectively process targets of different sizes by carrying out target detection under a plurality of different scales, and can detect on different levels of a network at the same time, and the precision and stability of a detection result are improved by feature fusion.
The communication module of the invention adopts a hidden communication technology based on a coding layer, and introduces a binary Hamming matrix embedding mode facing to the reduction of the modification quantity into a hidden communication system of the coding layer, the embedding mode can reduce the modification quantity of information embedding on carrier information, and simultaneously, an interleaving module is introduced to improve an extraction scheme, so that the negative effect of the embedding method on the reliability of secret information is weakened. The interleaving function is to convert the continuous errors brought by the embedding mode into random errors, thereby being beneficial to decoding of the convolution codes and increasing the reliability of secret information. The matrix quantization extraction is an extraction scheme which is proposed based on the embedding mode and can enable the extracted information to carry out soft decision decoding, so that the error rate of secret information can be further reduced.
Drawings
In order to more clearly illustrate the technical solutions of the present invention, the drawings that are needed in the embodiments are briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a diagram of an overall framework of an underwater target detection system based on a covert communication technique and deep learning in an embodiment of the invention;
FIG. 2 is a diagram of an overall framework of a network model of an image processing module in an embodiment of the invention;
FIG. 3 is a block diagram of an image acquisition module in an embodiment of the invention;
Fig. 4 is a system block diagram of a coding layer hidden communication scheme in an embodiment of the present invention.
Fig. 5 is a schematic view of an underwater target detection robot housing in an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
Example 1
The invention provides an underwater target detection method based on a hidden communication technology and deep learning, which comprises the following steps:
acquiring an image of a target object in real time;
Preprocessing an image, and extracting features of the preprocessed image by using a convolutional neural network model to obtain boundary frame coordinates, category labels and confidence scores of each target;
The information transmission method based on the hidden communication realizes the safe transmission of the boundary frame coordinates, the category labels, the confidence scores and the communication information of each target.
In the present embodiment, image data including a target object is collected according to the requirements of a target detection task. The data can be obtained from the public data set, or can be collected or purchased by itself. The data set is ensured to contain the diversity and representativeness of each class of targets.
The collected images are labeled using LabelImg or RectLabel labeling tools, labeling the location and class of each target. Common target labeling formats include Bounding Box (rectangular box) and Mask.
The marked data set is divided into a training set, a verification set and a test set. Typically 70% -80% of the data is used as training set, 10% -15% of the data is used as verification set, and the rest of the data is used as test set. Ensuring a relatively uniform distribution of the target categories for each dataset.
And a data enhancement technology is applied to the training set, so that the diversity and the number of the data sets are increased, and the generalization capability of the model is improved. Common data enhancement methods include random cropping, scaling, panning, rotation, brightness adjustment, color conversion, and the like.
The annotation and image data are converted into the format required for model training. For the YOLOv model, the common data format is Darknet format or COCO format. The data set may be converted to a specified format using a corresponding tool.
During training, the image data is pre-processed, e.g., normalized, scaled to a fixed size, etc. Thus, the training process can be accelerated, and the performance and stability of the model can be improved.
The prepared dataset is loaded into the model for training using the data loading tool DataLoader provided by the deep learning framework PyTorch.
The above is the data collection and preprocessing process followed by the training process for the specific object detection model.
A network structure of the YOLOv8 model is built using a deep learning framework PyTorch. The construction process includes a combination of a downsampling network, a feature extraction layer, and a detection head.
The model is weight initialized, and may be initialized randomly or with pre-trained weights.
A penalty function of the YOLOv model is defined. The penalty function of YOLOv typically consists of a location penalty, a classification penalty, and a confidence penalty. The method can be realized by using the existing loss functions, such as cross entropy and the like, and can be correspondingly adjusted according to task requirements.
The prepared training set and verification set data is loaded using DataLoader of data loading tool PyTorch.
An appropriate optimizer is selected, here an SGD optimizer is selected, and then the corresponding super parameters of learning rate, weight decay, momentum, etc. are set. These hyper-parameters can be debugged and optimized through experimentation.
In each training cycle, a batch of training data is input into the model for forward propagation calculations, and then for backward propagation to update the model parameters. Model weights are updated according to the value of the loss function.
The rate of model training may be controlled by way of learning rate decay, typically by gradually decreasing the learning rate during the training process.
Model weights during the training process are saved periodically for subsequent model evaluation and reasoning use.
And deploying the stored model into an operating system of the underwater robot, and detecting and using the underwater target.
The underwater target detection needs to use an image acquired by an underwater robot in real time. Firstly, preprocessing an image acquired by an underwater robot, wherein the preprocessing comprises adjustment of image size, normalization of pixel values and the like. These preprocessing operations help to improve the performance and accuracy of the model.
And performing feature extraction on the preprocessed image by using a Convolutional Neural Network (CNN) model. YOLOv8 employs Darknet53 as a feature extraction network, which consists of multiple convolutional layers, residual blocks, and convergence layers.
YOLOv8 a plurality of fine-grained feature layers (fine-grained features) are introduced on the basis of Darknet for extracting feature information of different scales. By fusing features of different levels, contextual information of objects of different sizes can be better captured.
YOLOv8 use multiple detection heads (detection heads) to predict target boxes of different scales. Each detection head is responsible for predicting a target of a particular scale and outputting a corresponding classification probability, bounding box position, and confidence score.
And decoding the prediction result output by the detection head to obtain the boundary frame coordinates, the category labels and the confidence scores of each target.
Since the same object may be detected by multiple bounding boxes, the prediction box is filtered using NMS algorithm in order to eliminate duplicate detection results. The NMS will select the most likely target box based on the confidence score and overlap of the predicted boxes.
And further processing the target frame processed by the NMS, such as setting a confidence threshold value, and filtering the target frame with low confidence.
Finally, a category label, location information, and confidence score including the detected target frame are output.
In this embodiment, the apparatus for capturing an image of a target object in real time mainly includes two major components: high definition digtal camera module and LED light filling device.
And (3) remotely controlling the underwater robot, acquiring images of the region to be detected, and temporarily storing the acquired images into a storage system of the underwater robot.
In this embodiment, the information transmission method based on covert communication realizes secure transmission of the boundary frame coordinates, class labels, confidence scores and communication information of each target, and mainly includes the following components: transmitter, receiver, sonar, signal processing unit and control unit. The data transmission and reception is mainly divided into two stages: a data transmission phase and a data reception phase.
In the data transmission phase, the transmitter acquires information processed by the image processing module to be transmitted, which is referred to herein as secret information, from the storage system of the underwater robot. Some of the information that can be publicly transmitted is referred to herein as public information as a carrier of such processed information.
The signal processing unit carries out convolution encoding operation on the public information to form carrier information; the secret information is also subjected to convolutional encoding operation, and then is subjected to interleaving processing through an interleaver. Finally, the secret information is embedded into the public information by using a matrix embedding mode based on binary Hamming codes, which is called as secret carrying information.
The secret information is encoded by the source into a digital code, a process also known as "data packing". In this process, data is added with a preamble and a guard interval to enhance reliability and robustness of the signal.
The encoded data is modulated by the signal source onto a carrier frequency having quadrature characteristics, a process also known as "signal modulation". The purpose of modulation is to allow data to be transmitted farther and more stably in the channel.
The modulated signal is sent out through sonar.
In the data receiving phase, a data signal from a transmitting end is received by a receiver. The receiver demodulates the received signal, demodulates the signal from the carrier frequency, and restores the signal to the original data code. The demodulated data code is decoded by the receiver, and the preamble and the synchronization code are removed to restore the original digital code or analog signal. The secret information of the transmitting end is obtained.
Firstly, matrix quantization extraction is carried out on secret information, so that soft decision Viterbi decoding can be carried out on the extracted information, and the error rate of the secret information is reduced.
And the result of matrix quantization extraction is used as the input of a Viterbi soft decision decoder after de-interleaving, and finally the secret information is decoded and recovered.
Example two
As shown in fig. 1, the invention also discloses an underwater target detection system based on the hidden communication technology and deep learning, comprising: the device comprises an image processing module, an image acquisition module and a hidden communication module;
the image acquisition module is used for acquiring an image of a target object in real time;
The image processing module is used for preprocessing the image, and extracting features of the preprocessed image by using a convolutional neural network model to obtain the boundary frame coordinates, the category labels and the confidence scores of each target;
The hidden communication module is used for realizing the safe transfer of the boundary frame coordinates, the category labels, the confidence scores and the communication information of each target based on the information transmission method of the hidden communication.
In this embodiment, as shown in fig. 2, for an image processing module (including a preprocessing unit and a feature extraction unit), image data including a target object is collected according to the requirements of a target detection task. The data can be obtained from the public data set, or can be collected or purchased by itself. The data set is ensured to contain the diversity and representativeness of each class of targets.
The collected images are labeled using LabelImg or RectLabel labeling tools, labeling the location and class of each target. Common target labeling formats include Bounding Box (rectangular box) and Mask.
The marked data set is divided into a training set, a verification set and a test set. Typically 70% -80% of the data is used as training set, 10% -15% of the data is used as verification set, and the rest of the data is used as test set. Ensuring a relatively uniform distribution of the target categories for each dataset.
And a data enhancement technology is applied to the training set, so that the diversity and the number of the data sets are increased, and the generalization capability of the model is improved. Common data enhancement methods include random cropping, scaling, panning, rotation, brightness adjustment, color conversion, and the like.
The annotation and image data are converted into the format required for model training. For the YOLOv model, the common data format is Darknet format or COCO format. The data set may be converted to a specified format using a corresponding tool.
During training, the image data is pre-processed, e.g., normalized, scaled to a fixed size, etc. Thus, the training process can be accelerated, and the performance and stability of the model can be improved.
The prepared dataset is loaded into the model for training using the data loading tool DataLoader provided by the deep learning framework PyTorch.
The above is the data collection and preprocessing process followed by the training process for the specific object detection model.
A network structure of the YOLOv8 model is built using a deep learning framework PyTorch. The construction process includes a combination of a downsampling network, a feature extraction layer, and a detection head.
The model is weight initialized, and may be initialized randomly or with pre-trained weights.
A penalty function of the YOLOv model is defined. The penalty function of YOLOv typically consists of a location penalty, a classification penalty, and a confidence penalty. The method can be realized by using the existing loss functions, such as cross entropy and the like, and can be correspondingly adjusted according to task requirements.
The prepared training set and verification set data is loaded using DataLoader of data loading tool PyTorch.
An appropriate optimizer is selected, here an SGD optimizer is selected, and then the corresponding super parameters of learning rate, weight decay, momentum, etc. are set. These hyper-parameters can be debugged and optimized through experimentation.
In each training cycle, a batch of training data is input into the model for forward propagation calculations, and then for backward propagation to update the model parameters. Model weights are updated according to the value of the loss function.
The rate of model training may be controlled by way of learning rate decay, typically by gradually decreasing the learning rate during the training process.
Model weights during the training process are saved periodically for subsequent model evaluation and reasoning use.
And deploying the stored model into an operating system of the underwater robot, and detecting and using the underwater target.
The underwater target detection needs to use an image acquired by an underwater robot in real time. Firstly, preprocessing an image acquired by an underwater robot, wherein the preprocessing comprises adjustment of image size, normalization of pixel values and the like. These preprocessing operations help to improve the performance and accuracy of the model.
And performing feature extraction on the preprocessed image by using a Convolutional Neural Network (CNN) model. YOLOv8 employs Darknet53 as a feature extraction network, which consists of multiple convolutional layers, residual blocks, and convergence layers.
YOLOv8 a plurality of fine-grained feature layers (fine-grained features) are introduced on the basis of Darknet for extracting feature information of different scales. By fusing features of different levels, contextual information of objects of different sizes can be better captured.
YOLOv8 use multiple detection heads (detection heads) to predict target boxes of different scales. Each detection head is responsible for predicting a target of a particular scale and outputting a corresponding classification probability, bounding box position, and confidence score.
And decoding the prediction result output by the detection head to obtain the boundary frame coordinates, the category labels and the confidence scores of each target.
Since the same object may be detected by multiple bounding boxes, the prediction box is filtered using NMS algorithm in order to eliminate duplicate detection results. The NMS will select the most likely target box based on the confidence score and overlap of the predicted boxes.
And further processing the target frame processed by the NMS, such as setting a confidence threshold value, and filtering the target frame with low confidence.
Finally, a category label, location information, and confidence score including the detected target frame are output.
In this embodiment, as shown in fig. 3, the apparatus for an image acquisition module mainly includes two major components: high definition digtal camera module and LED light filling device.
And (3) remotely controlling the underwater robot, acquiring images of the region to be detected, and temporarily storing the acquired images into a storage system of the underwater robot.
In the present embodiment, as shown in fig. 4, for the communication module, the following components are mainly included: transmitter, receiver, sonar, signal processing unit and control unit. The data transmission and reception is mainly divided into two stages: a data transmission phase and a data reception phase.
In the data transmission phase, the transmitter acquires information processed by the image processing module to be transmitted, which is referred to herein as secret information, from the storage system of the underwater robot. Some of the information that can be publicly transmitted is referred to herein as public information as a carrier of such processed information.
The signal processing unit carries out convolution encoding operation on the public information to form carrier information; the secret information is also subjected to convolutional encoding operation, and then is subjected to interleaving processing through an interleaver. Finally, the secret information is embedded into the public information by using a matrix embedding mode based on binary Hamming codes, which is called as secret carrying information.
The secret information is encoded by the source into a digital code, a process also known as "data packing". In this process, data is added with a preamble and a guard interval to enhance reliability and robustness of the signal.
The encoded data is modulated by the signal source onto a carrier frequency having quadrature characteristics, a process also known as "signal modulation". The purpose of modulation is to allow data to be transmitted farther and more stably in the channel.
The modulated signal is sent out through sonar.
In the data receiving phase, a data signal from a transmitting end is received by a receiver. The receiver demodulates the received signal, demodulates the signal from the carrier frequency, and restores the signal to the original data code. The demodulated data code is decoded by the receiver, and the preamble and the synchronization code are removed to restore the original digital code or analog signal. The secret information of the transmitting end is obtained.
Firstly, matrix quantization extraction is carried out on secret information, so that soft decision Viterbi decoding can be carried out on the extracted information, and the error rate of the secret information is reduced.
And the result of matrix quantization extraction is used as the input of a Viterbi soft decision decoder after de-interleaving, and finally the secret information is decoded and recovered.
The invention mainly combines the related technologies of the target detection field and the hidden communication field, and fuses the target detection field and the hidden communication field into the underwater robot. In the image processing module, a yolov-based object detection algorithm is adopted; in a communication module, an improved coding layer covert communication scheme is employed.
Compared with the prior art, the technical scheme of the invention has the following advantages:
the method can perform target detection at a higher frame rate and has higher real-time performance.
The method adopts a deeper network structure, introduces a strategy of multi-scale prediction and feature fusion, and has better detection precision.
Is insensitive to the change of the number of targets and can better detect multi-target images.
The interleaving is used in the hidden communication system of the coding layer, so that continuous errors in information extraction can be well dispersed.
The addition of matrix quantization extraction can further reduce the error rate of secret information
Example III
The invention also discloses an underwater target detection robot based on the hidden communication technology and the deep learning, and the robot applies the underwater target detection method based on the hidden communication technology and the deep learning. A schematic view of the underwater target detection robot housing is shown in fig. 5.
The above embodiments are merely illustrative of the preferred embodiments of the present invention, and the scope of the present invention is not limited thereto, but various modifications and improvements made by those skilled in the art to which the present invention pertains are made without departing from the spirit of the present invention, and all modifications and improvements fall within the scope of the present invention as defined in the appended claims.
Claims (8)
1. The underwater target detection method based on the hidden communication technology and the deep learning is characterized by comprising the following steps of:
acquiring an image of a target object in real time;
Preprocessing the image, and extracting features of the preprocessed image by using a convolutional neural network model to obtain boundary frame coordinates, category labels and confidence scores of each target;
The information transmission method based on the hidden communication realizes the safe transmission of the boundary frame coordinates, the category labels, the confidence scores and the communication information of each target;
The information transmission method based on the hidden communication, the method for realizing the safe transfer of the boundary frame coordinates, the category labels and the confidence scores of each target and the communication information comprises the following steps:
Performing convolutional encoding operation on the public information to form carrier information; carrying out convolutional encoding operation on the secret information, and carrying out interleaving treatment on the secret information through an interleaver; embedding secret information into public information by using a matrix embedding mode based on binary Hamming codes to obtain secret carrying information; modulating the secret information to obtain modulated signals and transmitting the modulated signals through sonar; wherein, the secret information is information processed by the image processing module;
Carrying out matrix quantization extraction on the secret information, de-interleaving the result of matrix quantization extraction as input of a Viterbi soft decision decoder, and decoding to recover the secret information;
The information transmission method based on the hidden communication realizes the safe transmission of the boundary frame coordinates, the category labels and the confidence scores of each target and the communication information, and mainly comprises the following components: a transmitter, a receiver, a sonar, a signal processing unit and a control unit; the data transmission and reception is mainly divided into two stages: a data transmission stage and a data reception stage;
In the data transmission stage, the transmitter acquires information which is to be transmitted and is processed by the image processing module from a storage system of the underwater robot, wherein the information is called secret information; the public transmission information is used as a carrier of the processed information and is called public information;
The signal processing unit carries out convolution encoding operation on the public information to form carrier information; the secret information is subjected to convolutional coding operation, and then is subjected to interleaving treatment through an interleaver; finally, embedding the secret information into the public information by using a matrix embedding mode based on binary Hamming codes, which is called as secret carrying information;
the secret information is encoded into digital codes by the signal source, and is called as data packing; the data is added with a preamble and a guard interval;
The coded data is modulated by a signal source onto a carrier frequency having orthogonal characteristics, referred to as "signal modulation";
the modulated signal is sent out through sonar;
In the data receiving stage, receiving a data signal from a transmitting end through a receiver; the receiver demodulates the received signal, demodulates the signal from the carrier frequency, and restores the signal into the original data code; the demodulated data code is decoded by a receiver, the lead code and the synchronous code are removed, and the original digital code or analog signal is recovered to obtain secret carrying information of a transmitting end;
Firstly, carrying out matrix quantization extraction on secret information, so that soft decision Viterbi decoding is carried out on the extracted information;
And the result of matrix quantization extraction is used as the input of a Viterbi soft decision decoder after de-interleaving, and finally the secret information is decoded and recovered.
2. The underwater target detection method based on the covert communication technology and the deep learning according to claim 1, wherein the method for acquiring the image of the target object in real time comprises the following steps:
And acquiring an image of the region to be detected by using the high-definition camera module to obtain an image of the target object.
3. The underwater target detection method based on the hidden communication technology and the deep learning according to claim 1, wherein the method for preprocessing the image and extracting features of the preprocessed image by using a convolutional neural network model to obtain the boundary frame coordinates, the category labels and the confidence scores of each target comprises the following steps:
Adjusting the size of the image of the target object and normalizing the pixel value;
based on YOLOv model, introducing a plurality of fine-grained feature layers on the basis of Darknet, extracting and fusing the feature information of different scales of the preprocessed image, and predicting target frames of different scales by using a plurality of detection heads to obtain the boundary frame coordinates, class labels and confidence scores of each target.
4. Underwater target detection system based on hidden communication technology and deep learning, which is characterized by comprising: the device comprises an image processing module, an image acquisition module and a hidden communication module;
the image acquisition module is used for acquiring an image of a target object in real time;
The image processing module is used for preprocessing the image, and extracting features of the preprocessed image by using a convolutional neural network model to obtain the boundary frame coordinates, the category labels and the confidence scores of each target;
The hidden communication module is used for realizing the safe transfer of the boundary frame coordinates, the category labels, the confidence scores and the communication information of each target based on the information transmission method of the hidden communication.
5. The underwater target detection system based on the covert communication technique and the deep learning of claim 4, wherein the image acquisition module comprises: the high-definition camera module and the LED light supplementing device;
The high-definition camera module is used for collecting images of the region to be detected and obtaining images of the target object;
The LED light supplementing device is used for assisting the high-definition camera module.
6. The underwater target detection system based on the covert communication technique and the deep learning of claim 4, wherein the image processing module comprises: a preprocessing unit and a feature extraction unit;
The preprocessing unit is used for adjusting the size of the image of the target object and normalizing the pixel value;
The feature extraction unit is used for introducing a plurality of fine-grained feature layers on the basis of a YOLOv model and Darknet, extracting and fusing feature information of different scales of the preprocessed image, and predicting target frames of different scales by using a plurality of detection heads to obtain the boundary frame coordinates, category labels and confidence scores of each target.
7. The underwater target detection system based on the covert communication technique and deep learning of claim 4, wherein the covert communication module comprises: a transmitter, a receiver, a sonar, a signal processing unit and a control unit;
The signal processing unit is used for performing convolution encoding operation on the public information to form carrier information; carrying out convolutional encoding operation on the secret information, and carrying out interleaving treatment on the secret information through an interleaver; embedding secret information into public information by using a matrix embedding mode based on binary Hamming codes to obtain secret carrying information; modulating the secret information to obtain a modulated signal; wherein, the secret information is information processed by the image processing module;
the transmitter is used for transmitting the secret carrying information;
The sonar is used for sending out the modulated signals;
The receiver is used for receiving the data signal from the transmitter;
The control unit is used for carrying out matrix quantization extraction on the secret information, de-interleaving the result of matrix quantization extraction, and then taking the result as the input of the Viterbi soft decision decoder, and decoding and recovering the secret information.
8. An underwater target detection robot based on a covert communication technology and deep learning, which is characterized in that the robot applies the underwater target detection method based on the covert communication technology and the deep learning as set forth in any one of claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311537443.XA CN117611983B (en) | 2023-11-17 | 2023-11-17 | Underwater target detection method and system based on hidden communication technology and deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311537443.XA CN117611983B (en) | 2023-11-17 | 2023-11-17 | Underwater target detection method and system based on hidden communication technology and deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117611983A CN117611983A (en) | 2024-02-27 |
| CN117611983B true CN117611983B (en) | 2024-07-02 |
Family
ID=89948832
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311537443.XA Active CN117611983B (en) | 2023-11-17 | 2023-11-17 | Underwater target detection method and system based on hidden communication technology and deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117611983B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106197543A (en) * | 2016-07-13 | 2016-12-07 | 北方爆破科技有限公司 | The wireless communication system of a kind of Underwater Acoustic Environment monitoring and method |
| CN114863263A (en) * | 2022-07-07 | 2022-08-05 | 鲁东大学 | Snakehead detection method for intra-class shielding based on cross-scale hierarchical feature fusion |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU3320700A (en) * | 1999-03-26 | 2000-10-16 | Wulich Wave Ltd. | Underwater communication method, device, and system utilizing a doppler frequency shift |
| CN107465486B (en) * | 2017-06-28 | 2020-07-28 | 中国船舶重工集团公司第七一五研究所 | Cooperative coding communication method suitable for underwater acoustic network |
| CN109597032B (en) * | 2019-01-31 | 2020-02-14 | 中国科学院深海科学与工程研究所 | Underwater acoustic positioning communication method |
| CN110518935B (en) * | 2019-09-18 | 2020-10-02 | 中国海洋大学 | MC-CDMA-based underwater acoustic communication system and PAPR (Peak to average Power ratio) suppression method |
| RU2724145C1 (en) * | 2019-10-16 | 2020-06-22 | Федеральное государственное бюджетное учреждение науки Институт проблем морских технологий Дальневосточного отделения Российской академии наук (ИПМТ ДВО РАН) | Hydroacoustic monitoring station of underwater situation |
| JP7178016B2 (en) * | 2021-03-01 | 2022-11-25 | アイテック株式会社 | Image processing device and its image processing method |
| CN115471746A (en) * | 2022-08-26 | 2022-12-13 | 中船航海科技有限责任公司 | Ship target identification detection method based on deep learning |
| CN116707662A (en) * | 2023-06-13 | 2023-09-05 | 曲阜师范大学 | MIMO-FMT-TR underwater acoustic communication system and method |
| CN116895012A (en) * | 2023-07-21 | 2023-10-17 | 广东电网有限责任公司 | Underwater image abnormal target identification method, system and equipment |
-
2023
- 2023-11-17 CN CN202311537443.XA patent/CN117611983B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106197543A (en) * | 2016-07-13 | 2016-12-07 | 北方爆破科技有限公司 | The wireless communication system of a kind of Underwater Acoustic Environment monitoring and method |
| CN114863263A (en) * | 2022-07-07 | 2022-08-05 | 鲁东大学 | Snakehead detection method for intra-class shielding based on cross-scale hierarchical feature fusion |
Non-Patent Citations (1)
| Title |
|---|
| 刘伟伟 等.《 基于线性码的隐写编码研究进展》.江苏科技大学学报(自然科学版).2015,第362-370页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117611983A (en) | 2024-02-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110225341B (en) | Task-driven code stream structured image coding method | |
| CN111079683B (en) | Remote sensing image cloud and snow detection method based on convolutional neural network | |
| CN112565777B (en) | Deep learning model-based video data transmission method, system, medium and device | |
| CN118279734B (en) | Underwater particulate matter and biological image in-situ acquisition method, medium and system | |
| CN117611983B (en) | Underwater target detection method and system based on hidden communication technology and deep learning | |
| CN116320538A (en) | Semantic communication transmission method and system for substation inspection image | |
| CN110837806B (en) | Indoor positioning method based on deep convolutional memory network | |
| CN113280820A (en) | Orchard visual navigation path extraction method and system based on neural network | |
| CN114743099A (en) | Method and device for identifying faults of train key components | |
| CN118196628A (en) | Enhanced semantic-location feature fusion network method, device and application for extracting backbone based on different pre-training features | |
| CN116503641A (en) | Small target detection method based on multi-scale feature fusion | |
| CN114332699B (en) | Road condition prediction method, device, equipment and storage medium | |
| CN115713695A (en) | 6G satellite image landslide detection method based on multistage Transformer network | |
| CN114998117A (en) | Port ship intelligent identification method and system combining image defogging and denoising | |
| CN114418821A (en) | Blind watermark processing method based on image frequency domain | |
| CN106548143A (en) | It is a kind of based on facial recognition techniques and the face identification system and method for memory function | |
| CN112803951A (en) | Method for reducing noise of communication system received signal, receiving end and communication system | |
| CN116109944B (en) | Satellite image cloud target extraction method based on deep learning network architecture | |
| CN117291787B (en) | Traceability method and system based on data watermark | |
| CN118784093A (en) | Underwater unidirectional broadcast communication method and system | |
| Liu et al. | ForgeryTTT: Zero-Shot Image Manipulation Localization with Test-Time Training | |
| CN118298834A (en) | Task-driven image semantic communication method, system and training method thereof | |
| Wang et al. | Advanced scene text recognition and application for dynamic driving scenes | |
| CN118840743A (en) | Semantic communication method, device and equipment for binocular three-dimensional target recognition | |
| Lin | Visual prompt-based learning for aerial image dehazing |
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 |