A multi-angle fusion fish identification system

Abstract

The invention discloses a multi-angle fusion fish identification system, comprising an acquisition module, which uses X-rays to scan a target, obtains a three-dimensional stereogram of a target spine, and is connected to a data processing module; a camera module, which shoots an appearance image of the target, obtains an appearance image of the target, and is connected to the data processing module; a data storage library, which stores appearance images of various types of fish and spine images of different fish, and is connected to the data processing module; the data processing module intercepts three views according to the three-dimensional stereogram of the target spine transmitted by the acquisition module, performs image processing on the intercepted three views respectively, and counts the number of the target spine; the invention combines the recognition results of the internal features of the fish with the recognition results of the external features of the fish, so that the identification and appraisal of the fish is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of fish identification, and in particular to a multi-angle fusion fish identification system. Background Art

[0002] Traditional fish static identification technology initially generally uses manual screening, through artificially designed features, and then selects classifiers based on these features to identify the target. This method is not only labor-intensive, but also inefficient, with limited extracted features and low classification accuracy. With the development of image processing and machine vision technology, from eigenvalues, correlation coefficients, neural networks, hierarchical classification, support vector machines, to the subsequent use of various forms of convolutional neural networks to identify fish species, new fish identification technologies can not only reduce labor intensity, but also improve accuracy.

[0003] In addition, early fish identification technology mainly focused on species identification, fish age judgment and fish grading. The basic identification implementation process is: input fish images, select fish features, build a classifier, and input the feature vector into the classifier for species identification. The classification methods include: discriminant analysis, BP (Back Propagation) neural network, Bayesian, contour matching method and support vector machine (SVM).

[0004] Since the last century, computer information technology has developed rapidly, and deep learning has made major breakthroughs in many fields such as computer vision. Applying the powerful capabilities and advantages of deep learning to the field of fish image recognition is a major innovation and improvement in fish image recognition. Fishes vary in shape and size, making it more complicated to identify them. Moreover, different species of the same type of fish have similar appearance, size, texture and other characteristics. Traditional monitoring and identification methods are difficult to obtain correct identification results.

[0005] At the same time, existing fish identification often obtains external images of fish, and when acquiring image features, it often also obtains external features. It is impossible to extract and identify the internal features of the fish. Due to the similarity of the appearance of fish, the accuracy of fish identification is low.

[0006] For example, Chinese patent CN202010663026.X discloses a method for marine fish image recognition based on deep learning. Using a method based on a deep learning network, the recognition of marine fish images can be realized. Even if there is no understanding of the biological ocean, various marine fish can be identified. However, this recognition method only obtains the external image of the fish, and the recognition accuracy is low. Summary of the invention

[0007] The present invention mainly solves the problem of low accuracy of fish identification in the existing technology; it provides a multi-angle fusion fish identification system, which combines the difference in the spine of fish and the difference in external images, and then integrates the recognition results of the two to identify and authenticate the fish, with high recognition accuracy.

[0008] The above technical problems of the present invention are mainly solved by the following technical solutions: a multi-angle fusion fish identification system, including an acquisition module, which uses X-rays to scan the target, obtains a three-dimensional stereogram of the target spine, and is connected to a data processing module; a camera module, which takes an appearance image of the target, obtains an appearance image of the target, and is connected to the data processing module; a data storage library, which stores appearance images of various types of fish and spine images of different fish, and is connected to the data processing module; the data processing module, according to the three-dimensional stereogram of the target spine transmitted by the acquisition module, performs three-view interception, performs image processing on the intercepted three views respectively and counts the number of the target spine, and classifies the image processing results of the three views into a front view identification subset A through a classifier. i , side view recognition subset G j , top view recognition subset K m And the vertebra number recognition result is recorded as the vertebra number recognition subset F n , identify the front view subset A i , side view recognition subset G j , top view recognition subset K m and the vertebrae number identification subset F n The intersection Q is obtained by intersecting, and the fish shape recognition result is obtained by using the RCNN algorithm after image processing according to the shape image transmitted by the camera module. The target fish is identified according to the intersection Q and the fish shape recognition result. The target fish is collected by the acquisition module to collect data on the vertebrae of the target fish, and the number of vertebrae of the target fish is obtained. After comparing with the number of fish vertebrae in the data repository, the fish species is preliminarily determined. Then, the fish are further distinguished after classification by image recognition and classifier, and the several distinguished results are compared to obtain the recognition result of the internal features of the fish. The shape image is collected by the camera module, and the fish shape image is recognized by the deep learning network algorithm to obtain the recognition result of the external features. The recognition results of the internal features are combined with the recognition results of the external features to make the identification of the fish more accurate.

[0009] Preferably, it also includes a 3D projector and a result display unit, wherein the 3D projector is used to display the three-dimensional stereoscopic image of the target spine acquired by the acquisition module in 3D projection, and the result display unit is used to display the appearance image of the target acquired by the camera module and the recognition result of the target fish by the data processing module, and the result display unit displays the recognition result by means of chart annotation or name display, and the 3D projector and the result display unit are both connected to the data processing module.

[0010] Preferably, the acquisition module includes: a fixed frame, which plays a supporting role; at least two X-ray sources are provided, which are used for cross-emitting X-rays, installed on the fixed frame, and connected to the data processing module; a detector, which is used to acquire projection data, installed on the fixed frame, and connected to the data processing module.

[0011] Preferably, it also includes a synchronization trigger and a timing module; the synchronization trigger is used to generate a synchronization trigger signal, which is used to control the X-ray source to start emitting X-rays at the same time and is connected to the timing module; the timing module performs timing according to the timing instructions transmitted by the data processing module.

[0012] Preferably, it also includes a delay module and an electronic switch; the electronic switch is connected in parallel with the delay module, and the control end of the electronic switch is connected to the data processing module; the delay module generates a delay time T, and after the electronic switch is closed, the delay module does not work, and after the electronic switch is opened, the delay module works to delay the synchronization trigger signal, so that the two X-ray sources emit X-rays at intervals of period T.

[0013] Preferably, a fish fixing device is also included, and the fish fixing device is used to fix the target fish when the acquisition module and the camera module are working. The fish fixing device includes a box, a fish head fixer arranged in the box, and a water circulation device arranged at the bottom of the box. The box is a transparent glass box. The fish head fixer is used to fix the fish. The water circulation device is used to provide a living environment for the fish. The water circulation device includes a water tank installed at the bottom of the box, a water outlet pipe, a water inlet pipe, a water pump and an external water tank. Activated carbon filter cotton is installed in the water tank. The water tank is connected to the water pump through the water inlet pipe, the water outlet pipe is connected to the water tank, and the water pump is arranged in the external water tank. The fish fixing device is arranged to fix the fish during the identification process, so that data collection and image acquisition are more convenient.

[0014] Preferably, the fish head fixer comprises an elliptical hollow part and fixing belts fixedly arranged on both sides, the fixing belts are elastic fixing belts, and the elliptical hollow part is used to clamp the fish mouth by stretching the fixing belt or to wrap the fish head by bending the fixing belt. The fish head is fixed by the fish head fixer, and then the whole fish is fixed, which is convenient for the collection module to collect the fish's vertebra data, and the fish is fixed by the fixing belt stretching and clamping the fish mouth, which is convenient for the camera module to capture the fish's appearance image.

[0015] Preferably, if the unique fish species cannot be identified based on the intersection Q and the fish shape recognition results, the data of the intersection Q is preprocessed and divided into a training set and a validation set to retrain the RCNN algorithm model, and the verified RCNN algorithm model is obtained, and the fish shape recognition results are substituted into the RCNN algorithm model to output the final identification result, so that the accuracy of fish identification is further improved.

[0016] The beneficial effects of the present invention are as follows: (1) data of the target fish's vertebrae are collected through the acquisition module to obtain the number of the target fish's vertebrae, which is then compared with the number of fish vertebrae in the data storage library to preliminarily determine the fish species, and then the fish are further distinguished through image recognition and classification by a classifier, and the several distinguished results are compared to obtain the recognition results of the fish's internal features, and the appearance image is collected through the camera module to perform fish appearance image recognition to obtain the recognition results of the external features, and the recognition results of the internal features are combined with the recognition results of the external features to make the identification of the fish more accurate; (2) The system has a simple structure and is easy to operate, and is highly practical for the identification of fish. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 4 is a structural block diagram of a fish identification system according to an embodiment of the present invention.

[0018] In the figure, 1 is a collection module, 2 is a camera module, 3 is a data processing module, 4 is a first data transmission module, 5 is a second data transmission module, 6 is a data storage library, and 7 is a display module. DETAILED DESCRIPTION

[0019] The technical solution of the present invention is further specifically described below through embodiments and in conjunction with the accompanying drawings.

[0020] Embodiment 1: A multi-angle fusion fish identification system, such as Figure 1As shown, it includes an acquisition module 1, which uses X-rays to scan the target to obtain a three-dimensional stereogram of the target spine, and is connected to a data processing module; a camera module 2, which takes an appearance image of the target to obtain an appearance image of the target, and is connected to the data processing module; a data storage library 6, which stores appearance images of various types of fish and spine images of different fish, and is connected to the data processing module; a data processing module 3, which intercepts three views according to the three-dimensional stereogram of the target spine transmitted by the acquisition module, performs image processing on the intercepted three views respectively and counts the number of the target spine, and classifies the image processing results of the three views into a front view recognition subset A through a classifier. i , side view recognition subset G j , top view recognition subset K m And the vertebra number recognition result is recorded as the vertebra number recognition subset F n , identify the front view subset A i , side view recognition subset G j , top view recognition subset K m and the vertebrae number identification subset F n The intersection Q is obtained by intersecting, and the fish shape recognition result is obtained by using the RCNN algorithm after image processing according to the shape image transmitted by the camera module, and the target fish is identified according to the intersection Q and the fish shape recognition result; the display module 7 is used to display the three-dimensional stereogram of the target spine acquired by the acquisition module, the shape image of the target acquired by the camera module, and the recognition result of the target fish by the data processing module, and the display module is connected to the data processing module.

[0021] The present invention is provided with a first data transmission module 4 and a second data transmission module 5. The acquisition module is connected to the data processing module via the first data transmission module, and the camera module transmits data to the data processing module via the second data transmission module.

[0022] The display module of the present invention includes a 3D projector and a result display unit, wherein the 3D projector is used to display the three-dimensional stereogram of the target vertebra acquired by the acquisition module in 3D projection, and the result display unit displays the recognition results in the form of chart annotation or name display.

[0023] If the unique fish species cannot be identified based on the intersection Q and the fish appearance recognition results, the data of the intersection Q is preprocessed and divided into a training set and a validation set to retrain the RCNN algorithm model to obtain a verified RCNN algorithm model. The fish appearance recognition results are substituted into the RCNN algorithm model to output the final identification result.

[0024] The present invention performs image processing on the three views of the spine and the fish appearance image in the data processing module. The image processing method adopted is an existing image processing method, and its main purpose is to extract the characteristic parts in the image to facilitate the subsequent classifier or deep learning algorithm to identify and authenticate the fish.

[0025] The acquisition module is a CT scanner, the data processing module and the data storage library are computers, the camera module uses a high-definition camera, the first data transmission module is an information transmission cable, and the second data transmission module includes a wired transmission module, a wireless transmission module or a mobile storage device, wherein the mobile storage device is a hard disk, an SD card or a TF card.

[0026] The CT scanner includes a fixed frame, an X-ray source, a detector and a power supply battery. The fixed frame includes a base for supporting the ground, a bearing installed on the base, and a turntable installed on the base through the bearing. At least two X-ray sources are provided. The two X-ray sources emit X-rays crosswise, are fixedly installed on the turntable at a certain angle, and are connected to the data processing module. The detector is used to obtain projection data after X-ray irradiation, is installed on the turntable, and is connected to the data processing module. The power supply battery is used to power the X-ray source and the detector.

[0027] Furthermore, the CT scanner of the present invention is also provided with a synchronization trigger and a timing module. The synchronization trigger is used to generate a synchronization trigger signal, which is used to control the two X-ray sources to start emitting X-rays at the same time, and is connected to the timing module; the timing module performs timing according to the timing instruction transmitted by the data processing module. The data processing module sends a timing start instruction to the timing module, and the timing module starts timing according to the instruction. The synchronization trigger is provided with a clock. When the timing of the timing module matches the clock, a synchronization trigger signal is generated and sent to the two X-ray sources. The two X-ray sources emit X-rays at the same time, so that the detector receives a complete vertebral projection image.

[0028] In order to further make the projection image received by the detector clearer, the present invention is provided with a delay module and an electronic switch; the electronic switch is connected in parallel with the delay module, and the control end of the electronic switch is connected to the data processing module; the delay module generates a delay time length T, and after the electronic switch is closed, the delay module does not work, and after the electronic switch is disconnected, the delay module works to delay the synchronization trigger signal, so that the two X-ray sources emit X-rays at intervals of period T. Both X-ray sources are connected to the delay module, and when the delay module corresponding to one of the X-ray sources works, the synchronization signal is delayed, so that the corresponding X-ray source emits X-rays after a period of T. During the delay process, the detector detects the projection data after the other X-ray source emits X-rays, and improves the projection image, so that the projection image is clearer.

[0029] Embodiment 2, a multi-angle fusion fish identification system, the difference between this embodiment and embodiment 1 is that this embodiment adds a fish fixing device, the fish fixing device is used to fix the target fish when the acquisition module and the camera module are working, the fish fixing device includes a box body, a fish head fixer arranged in the box body, and a water circulation device arranged at the bottom of the box body, the box body is a transparent glass box body, the fish head fixer is used to fix the fish, the water circulation device is used to provide a living environment for the fish, the water circulation device includes a water tank installed at the bottom of the box body, a water outlet pipe, a water inlet pipe, a water pump and an external water tank, activated carbon filter cotton is installed in the water tank, the water tank is connected to the water pump through the water inlet pipe, the water outlet pipe is connected to the water tank, and the water pump is arranged in the external water tank.

[0030] The fish head fixer comprises an elliptical hollow part and fixing belts fixedly arranged on both sides, the fixing belts are elastic fixing belts, and the elliptical hollow part is used to clamp the fish mouth by stretching the fixing belt or to wrap the fish head by bending the fixing belt. The rest of the structure is the same as that of the first embodiment.

[0031] The working principle of the present invention is as follows: a target fish to be identified is placed in a box; before the target fish is placed in the box, more than half of the box is filled with clean water through an external water tank and a water pump, so that the target fish can survive stably in the box; the head of the target fish is passed through the pores of the elliptical hollow part, and the fixing belt is stretched to both sides, so that the elliptical hollow is deformed and an extrusion force is generated under the stretching of the fixing belts on both sides to fix the fish head; at this time, the box can be pushed to the scanning area of ​​the CT scanner to obtain a three-dimensional stereogram of the spine of the target fish; the CT scanner transmits the obtained three-dimensional stereogram to the computer, and the computer obtains the three-dimensional stereogram according to the obtained three-dimensional image. The three-dimensional image is first used to identify the internal features of the fish to obtain the intersection Q, and then the fixing belt is loosened so that the elliptical hollow part covers the lower part of the fish head. The fixing belts on both sides are bent, and mutually attractive magnets are set on the folded surfaces of the fixing belts to make the fixing belts fit together and wrap the fish head to fix the fish head. The fixing belt is set to be transparent, and the appearance image of the target fish is photographed by a high-definition camera. The fish appearance image can be captured as complete and clear as possible, and the fish appearance image is transmitted to the computer to obtain the fish appearance recognition result. The elements in the intersection Q are compared with the fish appearance recognition result one by one to obtain accurate fish identification and appraisal results.

[0032] The above-described embodiment is only a preferred solution of the present invention and does not limit the present invention in any form. There are other variations and modifications without exceeding the technical solution described in the claims.

Claims

1. A multi-angle fusion fish identification system, characterized in that: include: The acquisition module uses X-rays to scan the target, obtains a three-dimensional image of the target spine, and is connected to the data processing module; The camera module takes a picture of the target's appearance, obtains the target's appearance image, and is connected to the data processing module; A data repository storing images of various types of fish shapes and images of the spines of different fish species, connected to the data processing module; The data processing module intercepts three views according to the three-dimensional stereogram of the target spine transmitted by the acquisition module, processes the intercepted three views respectively and counts the number of target spines, and classifies the image processing results of the three views into the front view recognition subset A through the classifier recognition. i , side view recognition subset G j , top view recognition subset K m And the vertebra number recognition result is recorded as the vertebra number recognition subset F n , identify the front view subset A i , side view recognition subset G j , top view recognition subset K m and the vertebrae number identification subset F n The intersection Q is obtained by intersecting, and the fish shape recognition result is obtained by using the RCNN algorithm after image processing according to the shape image transmitted by the camera module, and the target fish is identified according to the intersection Q and the fish shape recognition result; If the unique fish species cannot be identified based on the intersection Q and the fish appearance recognition results, the data of the intersection Q is preprocessed and divided into a training set and a validation set to retrain the RCNN algorithm model to obtain a verified RCNN algorithm model. The fish appearance recognition results are substituted into the RCNN algorithm model to output the final identification result.

2. The multi-angle fusion fish identification system according to claim 1 is characterized in that: It also includes a 3D projector and a result display unit. The 3D projector is used to 3D project and display the three-dimensional stereoscopic image of the target spine acquired by the acquisition module. The result display unit is used to display the appearance image of the target acquired by the camera module and the recognition result of the target fish by the data processing module. The result display unit uses chart annotation or name display to display the recognition result. The 3D projector and the result display unit are both connected to the data processing module.

3. A multi-angle fusion fish identification system according to claim 1 or 2, characterized in that: The acquisition module comprises: A fixed frame plays a supporting role; At least two X-ray sources are provided for cross-emitting X-rays, mounted on a fixed frame and connected to the data processing module; The detector is used to obtain projection data, is installed on a fixed frame, and is connected to the data processing module.

4. The multi-angle fusion fish identification system according to claim 3 is characterized in that: Also included are synchronization triggers and timing modules; The synchronization trigger is used to generate a synchronization trigger signal, which is used to control the X-ray sources to start emitting X-rays at the same time, and is connected to the timing module; The timing module performs timing according to the timing instruction transmitted by the data processing module.

5. The multi-angle fusion fish identification system according to claim 4 is characterized in that: It also includes a delay module and an electronic switch; The electronic switch is connected in parallel with the delay module, and the control end of the electronic switch is connected to the data processing module; the delay module generates a delay time T. After the electronic switch is closed, the delay module does not work. After the electronic switch is opened, the delay module works to delay the synchronization trigger signal, so that the two X-ray sources emit X-rays at intervals of period T.

6. The multi-angle fusion fish identification system according to claim 1, characterized in that: It also includes a fish fixing device, which is used to fix the target fish when the acquisition module and the camera module are working. The fish fixing device includes a box body, a fish head fixer arranged in the box body, and a water circulation device arranged at the bottom of the box body. The box body is a transparent glass box body. The fish head fixer is used to fix the fish. The water circulation device is used to provide a living environment for the fish. The water circulation device includes a water tank installed at the bottom of the box body, a water outlet pipe, a water inlet pipe, a water pump and an external water tank. Activated carbon filter cotton is installed in the water tank. The water tank is connected to the water pump through the water inlet pipe, the water outlet pipe is connected to the water tank, and the water pump is arranged in the external water tank.

7. The multi-angle fusion fish identification system according to claim 6, characterized in that: The fish head fixer comprises an elliptical hollow portion and fixing belts fixedly arranged on both sides, wherein the fixing belts are elastic fixing belts, and the elliptical hollow portion is used to clamp the fish mouth by stretching the fixing belts or to wrap the fish head by bending the fixing belts.

Images