CN114596480A - Yoov 5 optimization-based benthic organism target detection method and system - Google Patents
Yoov 5 optimization-based benthic organism target detection method and system Download PDFInfo
- Publication number
- CN114596480A CN114596480A CN202210095105.4A CN202210095105A CN114596480A CN 114596480 A CN114596480 A CN 114596480A CN 202210095105 A CN202210095105 A CN 202210095105A CN 114596480 A CN114596480 A CN 114596480A
- Authority
- CN
- China
- Prior art keywords
- module
- yolov5
- data set
- benthos
- detection model
- 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.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 92
- 238000005457 optimization Methods 0.000 title claims description 20
- 238000012549 training Methods 0.000 claims abstract description 47
- 238000012545 processing Methods 0.000 claims abstract description 44
- 238000000034 method Methods 0.000 claims description 22
- 238000002372 labelling Methods 0.000 claims description 12
- 238000005070 sampling Methods 0.000 claims description 12
- 230000004913 activation Effects 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 11
- 230000007246 mechanism Effects 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 6
- 238000012795 verification Methods 0.000 claims description 6
- 238000013145 classification model Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 241001474374 Blennius Species 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000000638 solvent extraction Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 11
- 238000000605 extraction Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 241000258957 Asteroidea Species 0.000 description 1
- 241000257465 Echinoidea Species 0.000 description 1
- 241000251511 Holothuroidea Species 0.000 description 1
- 241000237509 Patinopecten sp. Species 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 235000020637 scallop Nutrition 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- Y02A40/00—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
- Y02A40/80—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in fisheries management
- Y02A40/81—Aquaculture, e.g. of fish
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a yolov 5-based optimized benthos target detection method, which comprises the following steps: s1: collecting images containing a plurality of benthos in a public database, and processing to obtain an original data set; s2: performing ray processing and marine background object shielding data enhancement processing on the original data set; s3: optimizing the yolov5 model to obtain a detection model for detecting the benthos target; s4: training and learning the detection model in the step S3 by using the data set subjected to the data enhancement processing in the step S2 to obtain a trained detection model; s5: and detecting the submarine biological target by using the trained detection model. The invention realizes the rapid and accurate identification of various benthos, is applied to the automatic fishing of the benthos, improves the fishing efficiency, reduces the manpower and reduces the fishing operation risk.
Description
Technical Field
The invention relates to the field of underwater intelligent image recognition, in particular to a yolov5 optimization-based benthic organism target detection method and system.
Background
Benthos, such as sea cucumber, sea urchin, starfish, scallop and the like, has high edible and medicinal values. Traditional fishing often relies on the manpower to carry out manual fishing, needs very big manpower, and efficiency is lower, and the operation personnel have certain risk. In recent years, image intelligent identification and artificial intelligence technologies are rapidly developed, and the image detection technology is applied to underwater fishing robots and can automatically identify the positions and the types of fishing objects, so that intelligent automatic fishing is completed, the fishing efficiency is greatly improved, the labor cost is reduced, and the operation risk is reduced.
The prior art discloses a benthos identification and detection method based on a single-stage deep learning network, which comprises the following steps: extracting the characteristics of the benthos based on a convolutional neural network; calculating regression errors of a true value frame and a prediction frame by adopting a GIoU strategy; and (4) carrying out dimension clustering based on a K-means algorithm. The scheme realizes the extraction of effective characteristics of the benthos by using a convolution-based neural network method, effectively inhibits the interference of background noise to the benthos, simultaneously highlights the characteristics of the benthos, and is beneficial to improving the detection and identification precision of the benthos. By means of the method of the GIoU, the regression error between the prediction box and the truth box can be accurately calculated, and therefore the convolution weight of the whole detection system can be fully optimized. In the self-supervision ship feature learning stage, a feature extraction model is trained on unmarked ocean target data in a momentum comparison mode, a dictionary is maintained as a sample queue in the momentum comparison process, and a key encoder is updated in a momentum updating mode; in the supervised marine target detection stage, a fast RCNN model is adopted for marine target detection, and the fast RCNN model comprises a backbone network for feature extraction, a regional candidate network for generating a region of interest and a RoI Head network; the characteristics obtained in the self-supervision ship characteristic learning stage are adopted by the backbone network to extract network parameters for initialization, marine environment and ship-related prior knowledge are provided for a target detection model, and parameter fine adjustment is carried out on the backbone network while the model is trained.
The existing schemes do not carry out corresponding treatment aiming at the diversity and complexity of marine environment, and the generalization capability of the model is insufficient. Or the model training effect is not good under the conditions that the picture samples are few and the categories are unbalanced. Meanwhile, the existing model network structure has low performance, and the detection effect of the benthos target can be improved by improving the existing model network structure. The detection efficiency of the used model is low, and certain limitation exists in practical application.
Disclosure of Invention
The invention mainly aims to provide a yolov 5-based optimized benthos target detection method, which is used for correspondingly processing the diversity and complexity of marine environment, optimizing a model and improving the accuracy of benthos identification.
It is a further object of the present invention to provide a seabed biological target detection system optimized based on yolov 5.
In order to solve the technical problems, the technical scheme of the invention is as follows:
a seabed organism target detection method based on yolov5 optimization comprises the following steps:
s1: collecting images containing a plurality of benthos in a public database, and processing to obtain an original data set;
s2: performing data enhancement processing such as light processing, ocean background object shielding and the like on the original data set;
s3: optimizing the yolov5 model to obtain a detection model for detecting the benthos target
S4: training and learning the detection model of the step S3 by using the data set subjected to the data enhancement processing of the step S2 to obtain a trained detection model;
s5: and detecting the submarine biological target by using the trained detection model.
Preferably, the step S1 specifically includes the following steps:
s1.1: obtaining an image containing a plurality of benthos from a public database and a crawler;
s1.2: and screening and labeling the image to obtain an original data set.
Preferably, in step S1.2, labeling is performed by using a labelimg tool to obtain labeling information, where the labeling information includes object type in the image, object center coordinates x and y, and length and width information of an object frame, where the length and width information of the object frame is: the ratio of the length and width of the object frame to the length and width of the picture.
Preferably, the step S2 specifically includes the following steps:
by means of switching the background color of the picture, adjusting the light brightness, adding different background backgrounds of the submarine environment and increasing shielding of benthos for the target object, the diversity of the data set is increased, wherein the formula for adjusting the light brightness is as follows:
g(x，yj＝a*f(x，y)+b
where f (x, y) is the original value of pixel (x, y), a and b are adjustment factors, and g (x, y) is the adjusted pixel value.
The shielding scheme is as follows: the method comprises the steps of taking common ocean scenes such as stones, seaweed and reefs as shielding objects, randomly placing the shielding objects at the upper, lower, left, right and center corners or the center of a target object, and setting a shielding coverage range by adjusting the length and width of a shielding object picture.
Preferably, in the step S3, the yolov5 model is optimized, specifically:
1) optimize yolov5 network structure, increase branch and the network layer of upsampling in many places in original yolov5 network to connect the operation respectively to it, increase and detect the head, and increase corresponding 3 groups of anchors frames respectively for the detection layer that increases, the total 5 groups of detection heads of yolov5 structure after the optimization, can strengthen the detectability to little object and big object. The method specifically comprises the following steps: leading out a branch from the 3 rd layer in the original yolov5 network, and performing connection operation with an up-sampling network layer added subsequently; and adding a set of convolutional layers behind the 10 th layer in the original yolov5 network, and leading out branches to be connected with the subsequently added up-sampling layer.
2) Optimizing an original Conv module of yolov5, and modifying a convolution layer activation function in the Conv module into a Half _ Hardswish activation function, wherein the Half _ Hardswish activation function is as shown in the formula:
3) in the last layer of backbone of yolov5, a SE-HHslide layer module is added;
4) adding a CBAM attention mechanism module in the Conv module after the optimization in the step 2);
5) optimizing a BottleneckCSP module of yolov5, uniformly changing convolution layers in the module into a Conv module subjected to optimization in 2), changing an original two-branch structure into a three-branch structure, combining outputs of three branches to serve as inputs of the module, and mapping a feature map of the optimized BottleneckCSP module as follows:
y1＝ConvHHs(x)
y2＝Bottleneck(ConvHHs(x))
y2＝Bottleneck(ConvHHs(x))
BottleneckCSP_Cat(x)＝Concat(y1，y2，y3)
in the formula, the BottleneckCSP _ Cat () is the output of the optimized BottleneckCSP module.
Preferably, before the step S4 of training and learning the detection model in the step S3 by using the data set after the data enhancement processing in the step S2, the method further includes the steps of:
extracting a backhaul network part of the detection model in the step S3, establishing a classification model, and pre-training on a large-scale marine organism classification data set to obtain a pre-training weight;
the pre-training weights are loaded into the detection model of step S3.
Preferably, the step S4 specifically includes the following steps:
s4.1: setting up an environment required by the training of the detection model;
s4.2: dividing the data set subjected to the data enhancement processing in the step S2 to obtain a training set, a verification set and a test set;
s4.3: setting a hyper-parameter of network model training;
s4.4: and loading the divided data set into the detection model for training to obtain the trained detection model.
Preferably, in step S4.2, 60% of the data set after the data enhancement processing in step S2 is used as a training set, 20% is used as a verification set, and 20% is used as a test set.
Preferably, hierarchical sampling is performed during the partitioning of the data set in step S4.1.
A yolov 5-based optimized benthic organism target detection system comprising:
the system comprises an original data module, a data processing module and a data processing module, wherein the original data module collects images containing various benthos in an open database and processes the images to obtain an original data set;
the enhancement module is used for performing light ray processing and marine background object shielding data enhancement processing on the original data set;
the detection model module is used for optimizing the yolov5 model to obtain a detection model for detecting the benthos target;
the training module is used for training and learning the detection model in the step S3 by using the data set subjected to the data enhancement processing in the step S2 to obtain a trained detection model;
and the detection module is used for detecting the benthos target by utilizing the trained detection model.
Compared with the prior art, the technical scheme of the invention has the beneficial effects that:
(1) according to the invention, through a data enhancement method, the sample balance of the data set can be effectively carried out, and the influence of the sample imbalance on model training is reduced. Meanwhile, by switching the background color and light of the picture and adding different background backgrounds of the seabed environment for the target object, the diversity of the data set can be increased, and therefore the generalization capability of the model is improved.
(2) According to the invention, model optimization is carried out on yolov5, the structural logic of the underlying convolutional layer is mainly modified, the image feature learning capability of the model is improved, SE-HHslide and CBAM attention mechanisms are introduced into multiple places of a network structure, the channel learning capability and the space learning capability of the model are improved, and the accuracy of submarine biological target detection is further improved.
(3) According to the invention, the yolov5 shallow network and the subsequent up-sampling network layer are connected, the detection head is added, and 3 corresponding groups of smaller anchors are added for the detection layer, so that the detection capability of the model on small objects is improved.
(4) The method uses the large-scale marine organism category data set to pre-train the classification model of the backhaul model of the yolov5 optimization model, and is favorable for the learning ability of the model to extract the marine organism picture characteristics.
(5) According to the invention, the hyper-parameters of the yolov5 optimization model, such as the learning rate, the number of learning rounds, the number of pictures learned each time and the like, are adjusted, so that the training effect of the model can be improved.
Drawings
FIG. 1 is a schematic flow chart of the method of the present invention.
Fig. 2 is a schematic diagram of an optimized yolov5 network structure.
Fig. 3 is a schematic structural diagram of a ConvHHs module according to an embodiment.
FIG. 4 is a block diagram of an SE-HHslayer according to an embodiment.
FIG. 5 is a schematic structural diagram of the BottleneckCSP _ Cat module.
FIG. 6 is a block diagram of a system according to the present invention.
Detailed Description
The drawings are for illustrative purposes only and are not to be construed as limiting the patent;
for the purpose of better illustrating the embodiments, certain features of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product;
it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The technical solution of the present invention is further described below with reference to the accompanying drawings and examples.
Example 1
The embodiment provides a method for detecting a submarine organism target based on yolov5 optimization, which is shown in fig. 1 and comprises the following steps:
s1: collecting images containing a plurality of benthos in a public database, and processing to obtain an original data set;
s2: performing ray processing and marine background object shielding data enhancement processing on the original data set;
s3: optimizing the yolov5 model to obtain a detection model for detecting the benthos target;
s4: training and learning the detection model in the step S3 by using the data set subjected to the data enhancement processing in the step S2 to obtain a trained detection model;
s5: submarine biological target detection by using trained detection model
The step S1 of collecting images containing a plurality of benthos from the public database and processing the images to obtain a raw data set specifically includes the following steps:
s1.1: obtaining an image containing a plurality of benthos from a public database and a crawler;
s1.2: and screening and labeling the image to obtain an original data set.
In the step S1.2, labeling is performed by using a labellimg tool to obtain labeling information, where the labeling information includes object types, object center coordinates x and y, and length and width information of an object frame in the image, and the length and width information of the object frame is: the ratio of the length and width of the object frame to the length and width of the picture.
The step S2 of performing data enhancement processing on the original data set specifically includes the following steps:
by means of switching the background color of the picture, adjusting the light brightness, adding different background backgrounds of the submarine environment and increasing shielding of benthos for the target object, the diversity of the data set is increased, wherein the formula for adjusting the light brightness is as follows:
＝(x，y)＝a*f(x，y)+b
where f (x, y) is the original value of pixel (x, y), a and b are adjustment factors, and g (x, y) is the adjusted pixel value.
The shielding scheme is as follows: the method comprises the steps of taking a common ocean scene comprising stones, seaweed and reefs as shielding objects, randomly placing the shielding objects at the upper, lower, left, right and center corners or the center of a target object, and setting a shielding coverage range by adjusting the length and width of a shielding object picture.
After the data enhancement processing, the diversity of the data set can be increased, so that the generalization capability of the model is improved.
Before the step S4 of training and learning the detection model in step S3 by using the data set after the data enhancement processing in step S2, the method further includes the steps of:
extracting a backhaul network part of the detection model in the step S3, establishing a classification model, and pre-training on a large-scale marine organism classification data set to obtain a pre-training weight;
the pre-training weights are loaded into the detection model of step S3.
The obtained pre-training weight is used as the prior knowledge of the detection model, thereby being beneficial to the training and learning of the target detection model
The step S4 is to train and learn the detection model in the step S3 by using the data set after the data enhancement processing in the step S2, and specifically includes the following steps:
s4.1: setting up an environment required by the training of the detection model;
s4.2: dividing the data set subjected to the data enhancement processing in the step S2 to obtain a training set, a verification set and a test set;
s4.3: setting a hyper-parameter of network model training;
s4.4: and loading the divided data set into the detection model for training to obtain the trained detection model.
In the training mode of the model, a large-scale marine organism category data set is used for carrying out classification model pre-training on a backbone model of the yolov5 optimization model, and pre-training parameters serve as priori knowledge of a marine organism target detection model, so that the learning capacity of the model for extracting marine organism picture features is facilitated.
In step S4.2, 60% of the data set after the data enhancement processing in step S2 is used as a training set, 20% is used as a verification set, and 20% is used as a test set.
In step S4.2, hierarchical sampling is performed during the data set partitioning process, that is, the distribution of each category follows the above ratio, so as to avoid the data set distribution difference from being too large.
The partial hyper-parameters are set as: data set size per training: batch-size of 10; maximum number of training rounds: 300 parts of epochs; model learning rate lr0: 0.002; cyclic learning rate lrf: 0.15; the weight attenuation coefficient weight _ decay is 0.0003; preheating learning momentum warmup _ momentum is 0.8; the preheating initial learning rate warmup _ bias _ lr: 0.08.
Example 2
The embodiment provides specific contents for optimizing yolov5 on the basis of embodiment 1, and the specific contents comprise:
in the step S3, the yolov5 model is optimized, specifically:
1) the size of the image presented by the submarine organism target volume or due to the reasons of the distance of a camera and the like is diversified, the image size is smaller and larger, the model can better identify small objects in a shallow layer, and has stronger identification capability for large objects in a deep layer, so that branches and subsequent upper sampling network layers are added at a plurality of positions of a yolv 5 network, the branches and the subsequent upper sampling network layers are correspondingly connected, a detection head is added, 3 groups of corresponding anchors frames are added for the detection layer, the length and the width of the anchors frame added in the shallow layer are respectively set to be [5,6], [7,12], [16,10], the length and the width of the anchors frame added in the deep layer are respectively set to be [200,120], [520,110], [510,300], and the detection capability for the small objects and the large objects is improved; the method specifically comprises the following steps: leading out a branch from the 3 rd layer in the original yolov5 network, and performing connection operation with an up-sampling network layer added subsequently; adding a group of convolutional layers behind a 10 th layer in the original yolov5 network, and leading out branches to be connected with an up-sampling layer which is added subsequently; the optimized yolov5 network architecture is shown in fig. 2;
2) optimizing an original Conv module of yolov5, modifying a convolution layer activation function in the Conv module into a Half _ Hardswish activation function, and avoiding the problem that the original activation function cannot perform gradient descent optimization iterative training under the condition of a small input value, wherein the Half _ Hardswish activation function is as shown in the formula:
the optimized module is named as a ConvHHs module, and the structure diagram of the ConvHHs module is shown in figure 3;
3) in the last layer of the backbone of yolov5, an SE-HHslide layer module is added, wherein the SE-HHslide layer is an attention module and is mainly used for learning the weight of the layer channel, so that the sensitivity of the model to the channel characteristics is improved, and pixels of different channels obtain different importance. The SE-HHslayer module mainly obtains weight values of all positions of the characteristic diagram through learning the input characteristic diagram by a fully-connected neural network and a Half _ Hardswish and sigmoid activation function, and then obtains an output characteristic diagram by performing dot multiplication on the weight values and the input characteristic diagram, wherein the structure is shown in FIG. 4;
4) a CBAM attention mechanism module is added in a shallow ConvHHs module, the CBAM combines the attention mechanism of space and channel, not only measures the importance of pixels of different channels, but also measures the importance of pixels at different space positions in the same channel, and the feature extraction capability of a convolutional network can be improved;
and a SE-HHslayer and CBAM attention mechanism is introduced into a plurality of places of the network structure, so that the channel learning capacity and the space learning capacity of the model are improved, and the accuracy of detecting the submarine biological target is further improved.
5) Optimizing a BottleneckCSP module of yolov5, uniformly converting convolution layers in the module into a ConvHHs module, changing an original two-branch structure into a three-branch structure, combining outputs of three branches to be used as an input of the module, and naming the optimized module as a BottleneckCSP _ Cat module, wherein a characteristic diagram of the BottleneckCSP _ Cat module is mapped as follows:
y1＝ConvHHs(x)
y2＝Bottleneck(ConvHHs(x))
y2＝Bottleneck(ConvHHs(x))
BottleneckCSP_Cat(x)＝Concat(y1，y2，y3)
in the formula, the BottleneckCSP _ Cat () is the output of the optimized BottleneckCSP module.
The structure of the BottleneckCSP _ Cat module is shown in FIG. 5, and a superposition branching and convolution module is added to the BottleneckCSP in yolov5, so that the capability of extracting image features by convolution is improved.
Example 3
The present embodiment provides a seabed organism target detection system optimized based on yolov5, as shown in fig. 6, including:
the system comprises an original data module, a data processing module and a data processing module, wherein the original data module collects images containing various benthos in an open database and processes the images to obtain an original data set;
the enhancement module is used for carrying out enhancement processing on the original data set, including light processing and marine background object shielding data;
the detection model module is used for optimizing the yolov5 model to obtain a detection model for detecting the benthos target;
the training module is used for training and learning the detection model in the step S3 by using the data set subjected to the data enhancement processing in the step S2 to obtain a trained detection model;
and the detection module is used for detecting the benthos target by utilizing the trained detection model.
The same or similar reference numerals correspond to the same or similar parts;
the terms describing positional relationships in the drawings are for illustrative purposes only and are not to be construed as limiting the patent;
it should be understood that the above-described embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.
Claims (10)
1. A seabed organism target detection method based on yolov5 optimization is characterized by comprising the following steps:
s1: collecting images containing a plurality of benthos in a public database, and processing to obtain an original data set;
s2: performing ray processing and marine background object shielding data enhancement processing on the original data set;
s3: optimizing the yolov5 model to obtain a detection model for detecting the benthos target;
s4: training and learning the detection model in the step S3 by using the data set subjected to the data enhancement processing in the step S2 to obtain a trained detection model;
s5: and detecting the benthos target by using the trained detection model.
2. The yolov 5-based optimized benthic organism target detection method according to claim 1, wherein the step S1 comprises the following steps:
s1.1: obtaining an image containing a plurality of benthos from a public database and a crawler;
s1.2: and screening and labeling the image to obtain an original data set.
3. The method for detecting the benthos targets based on yolov5 optimization of claim 2, wherein in step S1.2, labeling is performed by using labelimg tool to obtain labeling information, and the labeling information includes object type in the image, object center coordinates x and y, and length and width information of an object frame, wherein the length and width information of the object frame is: the ratio of the length and width of the object frame to the length and width of the picture.
4. The yolov 5-based optimized benthic organism target detection method according to claim 1, wherein the step S2 comprises the following steps:
by means of switching picture background color, adjusting light brightness, adding different submarine environment backgrounds for target objects and increasing submarine biological shielding, diversity and complexity of a data set are increased, wherein the formula for adjusting light brightness is as follows:
g(x，y)＝a*f(x，y)+b
where f (x, y) is the original value of pixel (x, y), a and b are adjustment factors, and g (x, y) is the adjusted pixel value.
The shielding scheme is as follows: taking a common ocean scene comprising stones, seaweed and reefs as a shelter, randomly placing the shelter at the upper, lower, left, right and center corners or the center of a target object, and setting a shelter coverage range by adjusting the length and width of a shelter picture.
5. The method for detecting the benthos targets based on yolov5 optimization of claim 1, wherein the yolov5 model is optimized in step S3, and specifically comprises the following steps:
1) optimize yolov5 network structure, increase branch and upsampling network layer in many places in original yolov5 network to connect the operation respectively to it, increase and detect the head, and for the detection layer of increase respectively increase corresponding 3 groups of anchors frame, the total 5 groups of detection heads of yolov5 structure after optimizing can strengthen the detectability to little object and big object, specifically do: leading out a branch from the 3 rd layer in the original yolov5 network, and performing connection operation with an up-sampling network layer added subsequently; adding a group of convolutional layers behind the 10 th layer in the original yolov5 network, and leading out branches to be connected with the subsequently added upper sampling layers;
2) optimizing an original Conv module of yolov5, and modifying a convolution layer activation function in the Conv module into a Half _ Hardswish activation function, wherein the Half _ Hardswish activation function is as shown in the formula:
3) adding a SE-HHslayer module in the last layer of backbone of yolov 5;
4) adding a CBAM attention mechanism module in the Conv module after the optimization in the step 2);
5) optimizing the BottleneckCSP module of yolov5, uniformly converting the convolution layers in the module into Conv modules subjected to optimization in 2), changing the original two-branch structure into a three-branch structure, combining the outputs of the three branches as the input of the module, and mapping the feature map of the optimized BottleneckCSP module as follows:
y1＝ConvHHs(x)
y2＝Bottleneck(ConvHHs(x))
y2＝Bottleneck(ConvHHs(x))
BottleneckCSP_Cat(x)＝Concat(y1，y2，y3)
in the formula, the BottleneckCSP _ Cat () is the output of the optimized BottleneckCSP module.
6. The method for detecting marine benthos targets based on yolov5 optimization of claim 1, wherein before the step S4, the step S3 detection model is trained and learned by using the data set after the step S2 data enhancement, the method further comprises the following steps:
extracting a backhaul network part of the detection model in the step S3, establishing a classification model, and pre-training on a large-scale marine organism classification data set to obtain a pre-training weight;
the pre-training weights are loaded into the detection model of step S3.
7. The yolov 5-based optimized benthic organism target detection method according to claim 6, wherein the step S4 comprises the following steps:
s4.1: setting up an environment required by the training of the detection model;
s4.2: dividing the data set subjected to the data enhancement processing in the step S2 to obtain a training set, a verification set and a test set;
s4.3: setting a hyper-parameter of network model training;
s4.4: and loading the divided data set into the detection model for training to obtain the trained detection model.
8. The method for detecting marine organism targets based on yolov5 optimization of claim 7, wherein in step S4.2, 60% of the data set after the data enhancement processing of step S2 is used as a training set, 20% is used as a verification set, and 20% is used as a test set.
9. The yolov 5-based optimized benthic organism target detection method according to claim 8, wherein in step S4.1, hierarchical sampling is performed during data set partitioning.
10. A yolov 5-based optimized benthic organism target detection system, comprising:
the system comprises an original data module, a data processing module and a data processing module, wherein the original data module collects images containing various benthos in an open database and processes the images to obtain an original data set;
the enhancement module is used for performing light ray processing and marine background object shielding data enhancement processing on the original data set;
the detection model module is used for optimizing the yolov5 model to obtain a detection model for detecting the benthos target;
the training module is used for training and learning the detection model in the step S3 by using the data set subjected to the data enhancement processing in the step S2 to obtain a trained detection model;
and the detection module is used for detecting the benthos target by using the trained detection model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210095105.4A CN114596480A (en) | 2022-01-26 | 2022-01-26 | Yoov 5 optimization-based benthic organism target detection method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210095105.4A CN114596480A (en) | 2022-01-26 | 2022-01-26 | Yoov 5 optimization-based benthic organism target detection method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114596480A true CN114596480A (en) | 2022-06-07 |
Family
ID=81805546
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210095105.4A Pending CN114596480A (en) | 2022-01-26 | 2022-01-26 | Yoov 5 optimization-based benthic organism target detection method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114596480A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR102486499B1 (en) * | 2022-07-21 | 2023-01-10 | 한국해양과학기술원 | System for exploring tidal-flat burrows using image-based artificial intelligence object recognition and method for constructing tidal-flat ecological environment spatial information using therof |
| KR102486498B1 (en) * | 2022-07-21 | 2023-01-10 | 한국해양과학기술원 | System and method for detecting tidal-flat burrows and constructing tidal-flat ecological environment spatial information using image-based artificial intelligence object recognition |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113688948A (en) * | 2021-10-11 | 2021-11-23 | 中核检修有限公司 | Method for identifying attached marine organism species based on YOLO v5 |
| CN113901944A (en) * | 2021-10-25 | 2022-01-07 | 大连理工大学 | Marine organism target detection method based on improved YOLO algorithm |
| KR102351398B1 (en) * | 2021-06-17 | 2022-01-14 | (주)지오시스템리서치 | Method and system for detecting marine life using imaging equipment |
-
2022
- 2022-01-26 CN CN202210095105.4A patent/CN114596480A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR102351398B1 (en) * | 2021-06-17 | 2022-01-14 | (주)지오시스템리서치 | Method and system for detecting marine life using imaging equipment |
| CN113688948A (en) * | 2021-10-11 | 2021-11-23 | 中核检修有限公司 | Method for identifying attached marine organism species based on YOLO v5 |
| CN113901944A (en) * | 2021-10-25 | 2022-01-07 | 大连理工大学 | Marine organism target detection method based on improved YOLO algorithm |
Non-Patent Citations (1)
| Title |
|---|
| 毛国君等: "基于改进YOLO-V4 网络的浅海生物检测模型", 农业工程学报, vol. 37, no. 12, 30 June 2021 (2021-06-30) * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR102486499B1 (en) * | 2022-07-21 | 2023-01-10 | 한국해양과학기술원 | System for exploring tidal-flat burrows using image-based artificial intelligence object recognition and method for constructing tidal-flat ecological environment spatial information using therof |
| KR102486498B1 (en) * | 2022-07-21 | 2023-01-10 | 한국해양과학기술원 | System and method for detecting tidal-flat burrows and constructing tidal-flat ecological environment spatial information using image-based artificial intelligence object recognition |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111310862B (en) | Image enhancement-based deep neural network license plate positioning method in complex environment | |
| CN114170497B (en) | Multi-scale underwater fish school detection method based on attention module | |
| CN115082855B (en) | Pedestrian shielding detection method based on improved YOLOX algorithm | |
| CN112800838A (en) | Channel ship detection and identification method based on deep learning | |
| CN111967480A (en) | Multi-scale self-attention target detection method based on weight sharing | |
| CN110108704A (en) | A kind of automatic monitoring and pre-alarming method of cyanobacteria and its automatic monitoring and alarming system | |
| CN114596480A (en) | Yoov 5 optimization-based benthic organism target detection method and system | |
| Wang et al. | Deep learning-based visual detection of marine organisms: A survey | |
| CN114782982A (en) | Marine organism intelligent detection method based on deep learning | |
| CN113743417B (en) | Semantic segmentation method and semantic segmentation device | |
| CN110647802A (en) | Remote sensing image ship target detection method based on deep learning | |
| CN113221848B (en) | Hyperspectral open set field self-adaption method based on multi-classifier domain confrontation network | |
| Tang et al. | Pest-YOLO: Deep image mining and multi-feature fusion for real-time agriculture pest detection | |
| CN115205667A (en) | Dense target detection method based on YOLOv5s | |
| CN113743505A (en) | Improved SSD target detection method based on self-attention and feature fusion | |
| CN117079098A (en) | Space small target detection method based on position coding | |
| CN112084897A (en) | Rapid traffic large-scene vehicle target detection method of GS-SSD | |
| CN117333948A (en) | End-to-end multi-target broiler behavior identification method integrating space-time attention mechanism | |
| CN116168240A (en) | Arbitrary-direction dense ship target detection method based on attention enhancement | |
| CN113011436A (en) | Traditional Chinese medicine tongue color and fur color collaborative classification method based on convolutional neural network | |
| CN117152601A (en) | Underwater target detection method and system based on dynamic perception area routing | |
| CN114972869B (en) | Infrared weak target detection method based on anti-facts causal learning | |
| CN116452965A (en) | Underwater target detection and recognition method based on acousto-optic fusion | |
| Wang et al. | Underwater optical image object detection based on YOLOv7 algorithm | |
| CN115035429A (en) | Aerial photography target detection method based on composite backbone network and multiple measuring heads |
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 |