CN112966698A - Freshwater fish image real-time identification method based on lightweight convolutional network - Google Patents
Freshwater fish image real-time identification method based on lightweight convolutional network Download PDFInfo
- Publication number
- CN112966698A CN112966698A CN202110308905.5A CN202110308905A CN112966698A CN 112966698 A CN112966698 A CN 112966698A CN 202110308905 A CN202110308905 A CN 202110308905A CN 112966698 A CN112966698 A CN 112966698A
- Authority
- CN
- China
- Prior art keywords
- layer
- convolutional
- freshwater fish
- image
- network
- 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
- 241000251468 Actinopterygii Species 0.000 title claims abstract description 110
- 239000013505 freshwater Substances 0.000 title claims abstract description 84
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 26
- 238000012549 training Methods 0.000 claims abstract description 16
- 238000012360 testing method Methods 0.000 claims abstract description 8
- 238000011176 pooling Methods 0.000 claims description 33
- 238000007781 pre-processing Methods 0.000 claims description 7
- 238000005070 sampling Methods 0.000 claims description 6
- 238000011478 gradient descent method Methods 0.000 claims description 4
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 2
- 238000013519 translation Methods 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 abstract description 9
- 238000000605 extraction Methods 0.000 abstract description 9
- 238000004088 simulation Methods 0.000 description 7
- 238000002372 labelling Methods 0.000 description 5
- 241000252234 Hypophthalmichthys nobilis Species 0.000 description 4
- 230000007547 defect Effects 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 241001519451 Abramis brama Species 0.000 description 2
- 241001609213 Carassius carassius Species 0.000 description 2
- 241000252230 Ctenopharyngodon idella Species 0.000 description 2
- 241001275898 Mylopharyngodon piceus Species 0.000 description 2
- 238000009360 aquaculture Methods 0.000 description 2
- 244000144974 aquaculture Species 0.000 description 2
- 241001233037 catfish Species 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 201000004569 Blindness Diseases 0.000 description 1
- 241000252233 Cyprinus carpio Species 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 241000894007 species Species 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (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)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Farming Of Fish And Shellfish (AREA)
Abstract
The invention discloses a freshwater fish image real-time identification method based on a lightweight convolutional network, and aims to solve the problems of data set shortage, difficult feature extraction and low identification speed in the existing freshwater fish classification technology. The method comprises the following specific steps: (1) constructing a freshwater fish image data set; (2) constructing a lightweight deep convolutional neural network; (3) training a deep convolutional neural network; (4) identifying the freshwater fish test image in real time; the method automatically identifies the collected freshwater fish image in real time by constructing the freshwater fish image data set and utilizing the trained lightweight deep convolutional neural network, and has the advantages of no need of manually extracting fish body characteristics, high identification precision, high speed and low hardware resource consumption.
Description
Technical Field
The invention belongs to the technical field of image processing, and further relates to a freshwater fish image real-time identification method based on a lightweight convolutional network in the technical field of image identification. The method can be used for monitoring and identifying the captured fresh water fish species in real time in fishery monitoring, aquaculture and leisure fishing scenes, the identification result can be used for acquiring the fish species information, and a reference basis can be provided for the release of the rare fish species.
Background
In the traditional fish identification process, the fish body is mainly identified manually, the automatic freshwater fish identification technology can greatly reduce the labor intensity of workers, and the method can be applied to fishery monitoring, aquaculture and other aspects. In recent years, machine learning methods based on image features have been applied to fish image recognition, and have achieved good results. However, with the further development and the continuous deepening of the application degree of the machine learning technology, the following problems still exist in the field of fresh water fish image recognition: if different species of freshwater fishes generally have similar appearance, size and texture color, the distinguishing characteristics of the freshwater fishes cannot be accurately extracted by using the conventional freshwater fish identification method, so that the identification precision is not high; the identification method based on the convolutional neural network has the advantages that a data set is difficult to obtain, the existing method mostly depends on the existing marine fish data set, the freshwater fish cannot be identified, the complexity is high, and the algorithm cannot run in real time, so that deployment in embedded equipment is difficult. For example:
the patent document "a fish identification method, equipment and storage medium based on PCA" (patent application No. 201810118813.9, publication No. CN108460409A) applied by the university of agriculture in China proposes a fresh water fish identification method and device based on principal component analysis PCA. The method comprises the steps of constructing a horizontal and vertical coordinate matrix by extracting fish body outlines in fish images, carrying out principal component analysis according to the matrix to obtain principal components for distinguishing fishes, and finally completing fish identification according to the principal components for distinguishing fishes. The method disclosed by the patent application has the disadvantages that the method needs to automatically extract characteristic parameters such as the fish body contour and the like by using a program under a single background of relative structurization after the fish body leaves water, and the extraction and analysis processes of main components are complex, time-consuming and labor-consuming.
The patent document of Zhejiang agriculture and forestry university (patent application No. 201910912287.8, application publication No. CN110766013A) discloses a fish identification method based on a convolutional neural network. The method utilizes an ImageNet data set to pre-train a ResNet50 network, and utilizes a training set to optimize network parameters of a fish identification network to obtain a fish identification model. Compared with other fish identification methods, the method avoids the problem of insufficient subjectivity of manually extracted features in fish identification, and the training time is shortened and the higher accuracy is maintained by the training mode of initializing the weights of the convolution network through transfer learning. However, the method still has the following defects: the method has the advantages that the used network is complex, the parameter quantity is large, deployment of the method in the embedded equipment is difficult, the identification speed is low, most of fish species identified by the method are deep-sea fish due to the limitation of the existing data set, and the application value in the actual fresh water environment is not high.
Disclosure of Invention
The invention aims to provide a freshwater fish image real-time identification method based on a lightweight convolutional network aiming at overcoming the defects in the prior art, and aims to solve the problems of data set shortage, difficult feature extraction and slow identification speed in the prior freshwater fish identification technology.
In order to achieve the purpose, the idea of the method provided by the invention is as follows: in the fish species identification, a freshwater fish image data set is missing, so that large-scale training of a network cannot be supported, and a plurality of deep learning methods cannot be used; most of freshwater fishes are in irregular strips and have small fish targets, and some traditional freshwater fish identification methods have the defects of complex calculation, blindness and uncertainty in feature extraction and the like in the aspect of extracting image features, so that a lightweight convolutional neural network is built, and the image features of the freshwater fishes are automatically extracted by using the convolutional network with strong feature extraction capability; aiming at the problem of low identification speed of a convolutional neural network correlation method, the method reduces the number of network parameters as much as possible without reducing the identification accuracy so as to realize the real-time identification of the freshwater fish image. The method provided by the invention avoids manual feature extraction, ensures the identification accuracy, and can be applied and deployed in an actual scene without consuming a large amount of hardware resources.
The method comprises the following specific steps:
(1) constructing a freshwater fish image data set:
(1a) selecting at least 2700 freshwater fish images with the size of 244 multiplied by 244, wherein all the images at least cover 9 freshwater fish categories;
(1b) manually marking the type of the freshwater fish in each image, and marking the position of the freshwater fish in the image by using a rectangular bounding box;
(1c) preprocessing the marked image to obtain a training sample set;
(2) constructing a lightweight deep convolutional neural network:
(2a) a21-layer identification network is built, and the structure sequentially comprises the following steps: an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a fourth convolutional layer, a fourth pooling layer, a fifth convolutional layer, a fifth pooling layer, a sixth convolutional layer, a sixth pooling layer, a seventh convolutional layer, an eighth convolutional layer, a ninth convolutional layer, a tenth convolutional layer, a first output layer, an eleventh convolutional layer, an upsampling layer, a Concat splice layer, a twelfth convolutional layer, a thirteenth convolutional layer, and a second output layer; wherein the eleventh convolution layer is connected with the eighth convolution layer, and the Concat splice layer is connected with the fifth convolution layer;
(2b) the parameters of each layer are set as follows: the step sizes of the first to thirteenth convolutional layers are all set to be 1, the number of convolutional kernels in the first to thirteenth convolutional layers is respectively set to be 16, 32, 64, 128, 256, 512, 1024, 256, 512, 255, 128, 256, 255, the sizes of convolutional kernels in the eighth, tenth, eleventh, thirteenth convolutional layers are all set to be 1 x1, and the sizes of convolutional kernels in the rest convolutional layers are all set to be 3 x 3; all the pooling layers adopt a maximum pooling mode, the size of a pooling core is set to be 2 multiplied by 2, the step length of the sixth pooling layer is set to be 1, and the step lengths of the rest pooling layers are set to be 2; the up-sampling layer up-samples the feature map with the size of 13 multiplied by 128, and the output size is 26 multiplied by 128; the Concat splicing layer performs channel splicing on feature maps with the sizes of 13 multiplied by 128 and 13 multiplied by 256 respectively, and the output size is 26 multiplied by 384;
(3) training a deep convolutional neural network:
inputting the training sample set into a deep convolutional neural network, and iteratively updating the network parameters of each layer in the deep convolutional neural network by using a gradient descent method until a network loss function is converged to obtain a trained deep convolutional network;
(4) the method comprises the following steps of (1) identifying a freshwater fish test image in real time:
and (3) preprocessing the freshwater fish image to be recognized by adopting the same method as the step (1c), inputting the preprocessed freshwater fish image into a trained deep convolution neural network, and outputting the recognition result of the freshwater fish variety.
Compared with the prior art, the invention has the following advantages:
firstly, at least 2700 freshwater fish images with the size of 244 multiplied by 244 are constructed, all images at least cover freshwater fish image data sets of 9 freshwater fish categories, and all freshwater fish images are labeled manually, so that the problems that freshwater fish identification data sets are lost and a plurality of deep learning methods cannot be used in the prior art are solved, large-scale network training can be realized, and the range of freshwater fish variety identification is wider.
Secondly, the lightweight deep convolutional neural network is constructed, as the Concat splicing layer and the upper sampling layer are introduced into the network structure, the freshwater fish image to be identified is input into the trained network, the characteristics are extracted and spliced through the Concat splicing layer and the upper sampling layer, and the characteristics with multi-scale change are processed, so that the network characteristic extraction capability is enhanced, the network parameter quantity is optimized and reduced as far as possible while the identification accuracy is not reduced, the problems of difficult characteristic extraction and low identification speed of the deep learning method in the conventional method are solved, the freshwater fish image characteristics with high quality can be automatically extracted, and the freshwater fish image can be identified in real time.
Drawings
FIG. 1 is a flow chart of the present invention;
fig. 2 is a schematic diagram of a network structure constructed by the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
The implementation steps of the present invention are further described with reference to fig. 1.
Step 1, constructing a freshwater fish image data set.
Selecting at least 2700 freshwater fish images with the size of 244 multiplied by 244, wherein all the images at least cover 9 freshwater fish categories, and all the freshwater fish categories used by the method are all the current common freshwater fish varieties, namely bighead carp, silver carp, grass carp, black carp, crucian carp, bream, catfish and weever; the freshwater fish image data is mainly shot by a high-definition camera and acquired by a crawler technology, the shot data accounts for the main part, and the data acquired by the crawler technology is the secondary part.
Manually labeling the freshwater fish in each freshwater fish image, wherein the labeling type corresponds to the freshwater fish type, recording the vertex coordinates of each circumscribed rectangular frame used for labeling the freshwater fish and the type represented by the vertex coordinates, correspondingly generating a labeling file in an xml format for each picture, and then integrating and exporting all files in the xml format into a json labeling file.
And (3) preprocessing the marked image, namely sequentially carrying out random scaling, translation, rotation, mirror image and random cutting on the image, scaling the image to 224 multiplied by 224 space scale transformation, and obtaining the freshwater fish image data set. The data set was divided into 7: the ratio of 2 is randomly divided into a training set and a testing set.
And 2, constructing a lightweight deep convolutional neural network.
The lightweight deep convolutional neural network constructed by the present invention is further described with reference to fig. 2.
A21-layer identification network is built, and the structure sequentially comprises the following steps: an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a fourth convolutional layer, a fourth pooling layer, a fifth convolutional layer, a fifth pooling layer, a sixth convolutional layer, a sixth pooling layer, a seventh convolutional layer, an eighth convolutional layer, a ninth convolutional layer, a tenth convolutional layer, a first output layer, an eleventh convolutional layer, an upsampling layer, a Concat splice layer, a twelfth convolutional layer, a thirteenth convolutional layer, and a second output layer; wherein the eleventh convolution layer is connected to the eighth convolution layer and the Concat splice layer is connected to the fifth convolution layer.
The parameters of each layer are set as follows: the step sizes of the first to thirteenth convolutional layers are all set to be 1, the number of convolutional kernels in the first to thirteenth convolutional layers is respectively set to be 16, 32, 64, 128, 256, 512, 1024, 256, 512, 255, 128, 256, 255, the sizes of convolutional kernels in the eighth, tenth, eleventh, thirteenth convolutional layers are all set to be 1 x1, and the sizes of convolutional kernels in the rest convolutional layers are all set to be 3 x 3; all the pooling layers adopt a maximum pooling mode, the size of a pooling core is set to be 2 multiplied by 2, the step length of the sixth pooling layer is set to be 1, and the step lengths of the rest pooling layers are set to be 2; the up-sampling layer up-samples the feature map with the size of 13 multiplied by 128, and the output size is 26 multiplied by 128; the Concat concatenation layer performs channel concatenation on feature maps with sizes of 13 × 13 × 128 and 13 × 13 × 256, respectively, and outputs the feature maps with the sizes of 26 × 26 × 384.
In order to realize multi-scale joint prediction, the invention uses a Concat splicing layer to connect shallow features and deep features in a network structure, and the specific operation steps are as follows: the input image passes through the first 10 convolutional layers of the network to obtain the 1 st feature map, and the 1 st prediction is carried out; acquiring the output of the eighth convolution layer of the network structure, performing convolution and up-sampling operation on the result, splicing the acquired features with the fifth convolution features of the network structure to obtain a 2 nd feature map, and performing 2 nd prediction; the structure overcomes the defect of fish species identification in processing the problem of multi-scale change under the condition of adding extremely small calculation amount.
And 3, training the deep convolutional neural network.
Inputting the training sample set into the deep convolutional neural network, and iteratively updating the network parameters of each layer in the deep convolutional neural network by using a gradient descent method until the network loss function is converged to obtain the trained deep convolutional network.
The gradient descent method comprises the following steps:
in step 1, the learning rate of the convolutional network is set to 0.005.
And step 2, taking the difference value between the output value of the convolution network and the real label value of the image as a gradient value.
And 3, updating the weight of the convolutional neural network by using the following formula:
wherein the content of the first and second substances,
representing the updated weight of the convolutional neural network, ← representing the assignment operation, w representing the weight of the convolutional neural network randomly generating the weight subject to Gaussian distribution,
representing the gradient values of the convolutional neural network.
And 4, identifying the freshwater fish test image in real time.
And (3) preprocessing the freshwater fish image to be recognized by adopting the same method as the step (1c), inputting the preprocessed freshwater fish image into a trained deep convolution neural network, and outputting the recognition result of the freshwater fish variety.
In the actual use process, a display, a camera and a weight sensor are connected to the movable embedded development, a trained network model is deployed in an embedded development board, freshwater fish is placed on the weight sensor and in the field of view area of the camera, and the camera samples freshwater fish samples to obtain original images; the original image is identified by the method provided by the invention to obtain the identification result of the freshwater fish variety, and the weight and the identification result of the freshwater fish are displayed on a display.
The effects of the present invention will be further described with reference to comparative experiments.
1. Conditions of the experiment were simulated.
The hardware platform for simulation experiment operation of the invention is as follows: NVIDIA JETSON TX1 embedded platform: the CPU model is ARM Cortex-A57 MPCore (Quad-Core), the main frequency is 1.73GHz, and the memory size is 4 GB; the GPU is an NVIDIA Maxwell 256 core GPU. A CMOS camera: the Robotic C925e supports automatic zooming, inputs 1080P and 30 frame video streams, and accesses an experimental platform through a USB interface; a weight sensor: the QF-DLC485 high-precision digital sensor in the full-industry is connected into an experimental platform through an RS 485-USB cable.
The software platform for simulation experiment operation of the invention is as follows: the operating system is Ubuntu16.04 LTS, the OpenCV version is 3.2.0, and the version of Pytrch is 1.3.1.
2. And (5) simulating the contents of the experiment.
The simulation experiment of the invention is to adopt the method provided by the invention to identify the input freshwater fish image and obtain the identification result.
The self-built freshwater fish image data set used in the simulation experiment is mainly obtained by adopting high-definition camera shooting and a crawler technology, the shot data accounts for the main part, and the data obtained by the crawler technology accounts for the secondary part. The freshwater fish data set comprises 9 common data sets of bighead carp, silver carp, grass carp, black carp, common carp, crucian carp, bream, catfish and weever at present, wherein the data sets are all common freshwater fish types, the total number of 2700 images is 224 multiplied by 224, and the data sets are classified into 600 as a test set and 2100 as a training set.
And when the identification result in the simulation experiment is the same as the image label in the test set in the data set, the identification result of the freshwater fish is considered to be correct. And when the recognition result in the simulation experiment is different from the image label in the test set in the data set, the recognition result of the freshwater fish is considered to be incorrect.
In order to evaluate the effects of the present invention, the results of the simulation experiment identification using the method of the present invention were evaluated using the following evaluation indexes, and the calculation results are plotted in table 1:
TABLE 1 comparison of the present invention with other prior art
| Network parameter/MB | Average recognition time/s | Rate of identification accuracy |
| 33 | 0.08 | 91.6％ |
The above results show that: according to the invention, by constructing the freshwater fish image data set and utilizing the built and trained lightweight deep convolution network, the collected freshwater fish image can be automatically identified in real time without manually extracting fish body characteristics, the accuracy is high, the identification speed is high, the consumption of hardware resources is low, and meanwhile, the matched embedded equipment provides convenience for actual deployment. The method solves the problems of data set shortage, difficult feature extraction and slow recognition speed in the existing freshwater fish image recognition technology, and is a freshwater fish image real-time recognition method with strong practicability.
Claims (2)
1. A freshwater fish image real-time identification method based on a lightweight convolutional network is characterized in that a freshwater fish image is manually obtained and labeled to construct a data set, and the characteristics of the fish image are automatically extracted for identification by utilizing a built and trained lightweight convolutional neural network; the method comprises the following specific steps:
(1) constructing a freshwater fish image data set:
(1a) selecting at least 2700 freshwater fish images with the size of 244 multiplied by 244, wherein all the images at least cover 9 freshwater fish categories;
(1b) manually marking the type of the freshwater fish in each image, and marking the position of the freshwater fish in the image by using a rectangular bounding box;
(1c) preprocessing the marked image to obtain a training sample set;
(2) constructing a lightweight deep convolutional neural network:
(2a) a21-layer identification network is built, and the structure sequentially comprises the following steps: an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a fourth convolutional layer, a fourth pooling layer, a fifth convolutional layer, a fifth pooling layer, a sixth convolutional layer, a sixth pooling layer, a seventh convolutional layer, an eighth convolutional layer, a ninth convolutional layer, a tenth convolutional layer, a first output layer, an eleventh convolutional layer, an upsampling layer, a Concat splice layer, a twelfth convolutional layer, a thirteenth convolutional layer, and a second output layer; wherein the eleventh convolution layer is connected with the eighth convolution layer, and the Concat splice layer is connected with the fifth convolution layer;
(2b) the parameters of each layer are set as follows: the step sizes of the first to thirteenth convolutional layers are all set to be 1, the number of convolutional kernels in the first to thirteenth convolutional layers is respectively set to be 16, 32, 64, 128, 256, 512, 1024, 256, 512, 255, 128, 256, 255, the sizes of convolutional kernels in the eighth, tenth, eleventh, thirteenth convolutional layers are all set to be 1 x1, and the sizes of convolutional kernels in the rest convolutional layers are all set to be 3 x 3; all the pooling layers adopt a maximum pooling mode, the size of a pooling core is set to be 2 multiplied by 2, the step length of the sixth pooling layer is set to be 1, and the step lengths of the rest pooling layers are set to be 2; the up-sampling layer up-samples the feature map with the size of 13 multiplied by 128, and the output size is 26 multiplied by 128; the Concat splicing layer performs channel splicing on feature maps with the sizes of 13 multiplied by 128 and 13 multiplied by 256 respectively, and the output size is 26 multiplied by 384;
(3) training a deep convolutional neural network:
inputting the training sample set into a deep convolutional neural network, and iteratively updating the network parameters of each layer in the deep convolutional neural network by using a gradient descent method until a network loss function is converged to obtain a trained deep convolutional network;
(4) the method comprises the following steps of (1) identifying a freshwater fish test image in real time:
and (3) preprocessing the freshwater fish image to be recognized by adopting the same method as the step (1c), inputting the preprocessed freshwater fish image into a trained deep convolution neural network, and outputting the recognition result of the freshwater fish variety.
2. The method for identifying the freshwater fish image in real time based on the lightweight convolutional network as claimed in claim 1, wherein the preprocessing of the annotated image in the step (1c) is to perform random scaling, translation, rotation, mirroring, random cropping and scaling to 224 x 224 spatial scale transformation in sequence on the annotated image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110308905.5A CN112966698A (en) | 2021-03-23 | 2021-03-23 | Freshwater fish image real-time identification method based on lightweight convolutional network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110308905.5A CN112966698A (en) | 2021-03-23 | 2021-03-23 | Freshwater fish image real-time identification method based on lightweight convolutional network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112966698A true CN112966698A (en) | 2021-06-15 |
Family
ID=76278108
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110308905.5A Pending CN112966698A (en) | 2021-03-23 | 2021-03-23 | Freshwater fish image real-time identification method based on lightweight convolutional network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112966698A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114419364A (en) * | 2021-12-24 | 2022-04-29 | 华南农业大学 | Intelligent fish sorting method and system based on deep feature fusion |
| CN116071778A (en) * | 2023-03-31 | 2023-05-05 | 成都运荔枝科技有限公司 | Cold chain food warehouse management method |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109117877A (en) * | 2018-08-02 | 2019-01-01 | 南京师范大学 | A kind of Pelteobagrus fulvidraco and its intercropping kind recognition methods generating confrontation network based on depth convolution |
| CN109543679A (en) * | 2018-11-16 | 2019-03-29 | 南京师范大学 | A kind of dead fish recognition methods and early warning system based on depth convolutional neural networks |
| CN110516561A (en) * | 2019-08-05 | 2019-11-29 | 西安电子科技大学 | SAR image target recognition method based on DCGAN and CNN |
| CN110766013A (en) * | 2019-09-25 | 2020-02-07 | 浙江农林大学 | Fish identification method and device based on convolutional neural network |
| CN112287913A (en) * | 2020-12-25 | 2021-01-29 | 浙江渔生泰科技有限公司 | Intelligent supervisory system for fish video identification |
| CN112464744A (en) * | 2020-11-09 | 2021-03-09 | 湖北省农业科学院农产品加工与核农技术研究所 | Fish posture identification method |
-
2021
- 2021-03-23 CN CN202110308905.5A patent/CN112966698A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109117877A (en) * | 2018-08-02 | 2019-01-01 | 南京师范大学 | A kind of Pelteobagrus fulvidraco and its intercropping kind recognition methods generating confrontation network based on depth convolution |
| CN109543679A (en) * | 2018-11-16 | 2019-03-29 | 南京师范大学 | A kind of dead fish recognition methods and early warning system based on depth convolutional neural networks |
| CN110516561A (en) * | 2019-08-05 | 2019-11-29 | 西安电子科技大学 | SAR image target recognition method based on DCGAN and CNN |
| CN110766013A (en) * | 2019-09-25 | 2020-02-07 | 浙江农林大学 | Fish identification method and device based on convolutional neural network |
| CN112464744A (en) * | 2020-11-09 | 2021-03-09 | 湖北省农业科学院农产品加工与核农技术研究所 | Fish posture identification method |
| CN112287913A (en) * | 2020-12-25 | 2021-01-29 | 浙江渔生泰科技有限公司 | Intelligent supervisory system for fish video identification |
Non-Patent Citations (1)
| Title |
|---|
| 董洪义, 北京机械工业出版社 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114419364A (en) * | 2021-12-24 | 2022-04-29 | 华南农业大学 | Intelligent fish sorting method and system based on deep feature fusion |
| CN116071778A (en) * | 2023-03-31 | 2023-05-05 | 成都运荔枝科技有限公司 | Cold chain food warehouse management method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109816725B (en) | Monocular camera object pose estimation method and device based on deep learning | |
| CN109509187B (en) | Efficient inspection algorithm for small defects in large-resolution cloth images | |
| CN113065558A (en) | Lightweight small target detection method combined with attention mechanism | |
| CN109409365A (en) | It is a kind of that method is identified and positioned to fruit-picking based on depth targets detection | |
| CN108596102B (en) | RGB-D-based indoor scene object segmentation classifier construction method | |
| CN110163798B (en) | Method and system for detecting damage of purse net in fishing ground | |
| CN115082815B (en) | Tea bud picking point positioning method and device based on machine vision and picking system | |
| CN112967255A (en) | Shield segment defect type identification and positioning system and method based on deep learning | |
| CN111178177A (en) | Cucumber disease identification method based on convolutional neural network | |
| CN112966698A (en) | Freshwater fish image real-time identification method based on lightweight convolutional network | |
| CN110097537A (en) | A kind of meat quantitative analysis evaluation method based on three-D grain feature | |
| CN111696150A (en) | Method for measuring phenotypic data of channel catfish | |
| CN112215217B (en) | Digital image recognition method and device for simulating doctor to read film | |
| CN112183448B (en) | Method for dividing pod-removed soybean image based on three-level classification and multi-scale FCN | |
| Liu et al. | Deep learning based research on quality classification of shiitake mushrooms | |
| CN108932474B (en) | Remote sensing image cloud judgment method based on full convolution neural network composite characteristics | |
| CN114972646B (en) | Method and system for extracting and modifying independent ground objects of live-action three-dimensional model | |
| CN115099297A (en) | Soybean plant phenotype data statistical method based on improved YOLO v5 model | |
| CN114140665A (en) | Dense small target detection method based on improved YOLOv5 | |
| CN116310548A (en) | Method for detecting invasive plant seeds in imported seed products | |
| CN113435254A (en) | Sentinel second image-based farmland deep learning extraction method | |
| CN116295022A (en) | Pig body ruler measurement method based on deep learning multi-parameter fusion | |
| CN115719445A (en) | Seafood identification method based on deep learning and raspberry type 4B module | |
| CN114120359A (en) | Method for measuring body size of group-fed pigs based on stacked hourglass network | |
| CN113793385A (en) | Method and device for positioning fish head and fish tail |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210615 |