Automatic identification and accurate measurement equipment for fishes

Abstract

The invention discloses a fish automatic identification and precise measurement device, which comprises a data acquisition and input device, the data acquisition and input device comprises an electronic scale, a fish tray is laid on the top of the electronic scale, scale strips are arranged on the top of the fish tray, a support is mounted on the back of the electronic scale, and a fish storage box is mounted on the support. An industrial camera is fixedly connected to the side, away from the electronic scale, of the support and can shoot fishes placed on the surface of the fish tray, and the data acquisition and input device further comprises a computer. By arranging the integrated electronic scale, the industrial camera with the scale tray and the computer system, automatic synchronous acquisition of fish weight and form data is achieved, the problem that traditional manual measurement is low in efficiency is solved, the definition of image capture and the reliability of data transmission are guaranteed through the global shutter mmOS and double-channel transmission design, and the system is suitable for large-scale popularization and application. And a complete data basis is provided for subsequent intelligent analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of fishery resource investigation, and in particular to an automatic identification and precise measurement device for fish. Background Art

[0002] Fishery resource survey is an activity to scientifically evaluate the species, quantity, distribution, ecological characteristics, etc. of fish and other aquatic biological resources in oceans, rivers, lakes and other water bodies.

[0003] For example, the patent application number published on the China Patent Network is 201721817210.5, and the patent title is: "A composite measuring device for measuring fish body length and weight, comprising a measuring platform and a telescopic rod. The outer surface of the measuring platform is fixedly mounted with a waterproof housing. A support rod is provided parallel to the lower end of the measuring platform, and a rubber base is fixedly mounted at the lower end of the support rod. A measuring pressure plate is provided above the measuring platform. A connecting column is provided at the lower end of the measuring pressure plate, parallel to the upper end of the measuring platform, and a fish sample holding groove is fixedly mounted on the upper surface of the measuring pressure plate. In fishery resource surveys, three indicators of fish resources need to be measured: fish identification, fish body length, total length, and weight data. Existing technologies rely on the experience of surveyors to identify fish and manually measure physical length and weight. This results in slow measurement speed, time-consuming and labor-intensive work when the catch is large, and a high error rate. Especially in field environments, environmental influences reduce work efficiency and accuracy.

[0004] Therefore, it is necessary to design and create measurement and fish identification instruments for fishery resource surveys. Summary of the Invention

[0005] In order to solve the problems raised in the above-mentioned background technology, the purpose of the present invention is to provide an automatic identification and recognition and precise measurement equipment for fish, which has the advantage of improving the efficiency of fish measurement and identification, and solves the problem that the existing technology relies on the experience of surveyors to identify fish, and manually measures the physical length and weight, which has a slow measurement speed, is time-consuming and labor-intensive when the catch is relatively large, and has a large error rate, especially in the wild environment, where the work efficiency is affected by the environment.

[0006] To achieve the above-mentioned object, the present invention provides the following technical solutions: an automatic identification and precise measurement device for fish, comprising a data acquisition and input device; The data acquisition and input device includes an electronic scale, a fish tray is placed on the top of the electronic scale, a scale bar is opened on the top of the fish tray, a bracket is installed on the back of the electronic scale, an industrial camera is fixedly connected to the side of the bracket away from the electronic scale, and the industrial camera can take pictures of the fish placed on the surface of the fish tray. The data acquisition and input device also includes a computer, the computer and the industrial camera are interconnected through a data connection module, and a fish recognition model runs inside the computer.

[0007] As a preferred embodiment of the present invention, the deep learning model includes a data preprocessing module, the output end of the data preprocessing module is bidirectionally electrically connected to a feature extraction module, the output end of the feature extraction module is bidirectionally electrically connected to a model training module, the output end of the model training module is bidirectionally electrically connected to a model deployment module, and the output end of the model deployment module is bidirectionally electrically connected to the input end of a computer.

[0008] As a preferred embodiment of the present invention, the data connection module is composed of a USB connection cable or Ethernet data.

[0009] As a preferred embodiment of the present invention, the data preprocessing module is composed of a background removal unit, an image cropping and scaling unit, and a geometric calibration unit; the feature extraction module is composed of a multidimensional convolution unit, texture shape features, and color features; and the model training module is composed of a GPU server, a deep learning framework, and a verification and testing module.

[0010] As a preferred embodiment of the present invention, the image recognition model includes a reference scale establishment module, the input end of the reference scale establishment module is bidirectionally electrically connected to the output end of the feature extraction module, the output end of the reference scale establishment module is bidirectionally electrically connected to the spatial geometry calculation module, the output end of the spatial geometry calculation module is bidirectionally electrically connected to the feature storage and indexing module, the output end of the feature storage and indexing module is bidirectionally electrically connected to the feature matching module, the output end of the feature matching module is bidirectionally electrically connected to the geometry verification unit, the output end of the geometry verification unit is bidirectionally electrically connected to the recognition and retrieval decision module, and the output end of the recognition and retrieval decision module is bidirectionally electrically connected to the input end of the feedback module.

[0011] As a preferred embodiment of the present invention, the fish identification model also includes an output and application module, the output end of the computer is bidirectionally electrically connected to the output and application module, the output end of the output and application module is bidirectionally electrically connected to the classification storage module, and the output end of the classification storage module is bidirectionally electrically connected to the input end of the feature storage and indexing module.

[0012] As a preferred embodiment of the present invention, the output and application module is composed of type recognition results, length, weight measurement values ​​and statistical analysis reports, and the classification storage module is composed of original image data and annotation data.

[0013] As a preferred embodiment of the present invention, the feature matching module is composed of an approximate nearest neighbor search unit and a similarity measurement matching unit.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention physically integrates an electronic scale with a tray with graduated bars, and uses a bracket-mounted industrial camera to simultaneously capture the weight and panoramic image (including standard size references) of the fish as soon as it is placed. This eliminates the operational delays of manual item-by-item measurement and reduces the single-sample data acquisition time from the traditional 90 seconds to 5 seconds. The computer's built-in dual-model architecture enables parallel processing of species identification (deep learning) and body length / weight analysis (image recognition), improving efficiency by 200% compared to serial processing.

[0015] 2. The present invention forms a closed loop of preprocessing, feature extraction and model training through a modular processing flow. It uses U-Net network segmentation and adaptive histogram equalization processing to effectively eliminate interference such as water reflection and enhance key features such as fish fin texture, thereby improving the quality of subsequent model training data by about 40% and significantly improving recognition accuracy.

[0016] 3. By adopting USB / Ethernet dual-channel transmission, the present invention supports millisecond-level stable transmission of 20-megapixel images (packet loss rate ≤ 0.01%), which is particularly suitable for the bumpy environment of outdoor fishing boats. Compared with the single WiFi solution, the data integrity is improved by 99.2%.

[0017] 4. This invention breaks through traditional limitations through a multi-scale feature extraction architecture. The Inception module's multi-level perception of dorsal fin morphology increases the accuracy of distinguishing similar fish species (such as closely related species of Perciformes) from 72% to 89%, while the Transformer layer reduces the parsing error of body color patterns to 5 pixels.

[0018] 5. The present invention uses a reference scale establishment module to convert the actual size based on the pixel spacing of the scale bar, which improves the accuracy of manual reading by 20 times. The spatial geometry calculation module automatically calculates the fork length / body length by detecting the key points of the fish body.

[0019] 6. The present invention realizes data value conversion through an intelligent output system. The automatically generated statistical analysis report contains 12 ecological indicators and supports CSV / PDF dual-format output, which shortens the preparation time of field survey reports from 4 hours to 30 minutes, and the data traceability reaches 100%. At the same time, the data reflux mechanism of feature indexing archives the original images and annotation data of new samples to the feature library in real time, supports hourly online update of the model, and shortens the iteration cycle of the new species identification model from 30 days to 3 days.

[0020] 7. This invention forms a closed data loop through classified storage and feedback systems. The structured storage of raw images and annotated data reduces the model iteration cycle by 50%, supports incremental learning, and only requires 30% additional training to achieve the same recognition accuracy when adding new fish species. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a schematic diagram of the structure of the present invention; Figure 2 Schematic diagram of the system of the present invention.

[0022] In the figure: 1. Electronic scale; 2. Fish tray; 3. Scale bar; 4. Bracket; 5. Industrial camera; 6. Computer; 7. Data connection module. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0024] like Figures 1 to 2 As shown, the present invention provides an automatic identification and precise measurement device for fish, including a data acquisition and input device; The data acquisition and input device includes an electronic scale 1, a fish tray 2 is laid on the top of the electronic scale 1, a scale bar 3 is opened on the top of the fish tray 2, a bracket 4 is installed on the back of the electronic scale 1, and an industrial camera 5 is fixedly connected to the side of the bracket 4 away from the electronic scale 1. The industrial camera 5 can take pictures of the fish placed on the surface of the fish tray 2. The data acquisition and input device also includes a computer 6. The computer 6 and the industrial camera 5 are interconnected through a data connection module 7. A fish recognition model runs inside the computer 6. The fish recognition model consists of a deep learning model and an image recognition model. The deep learning model can identify and judge the fish species, and the image recognition model can identify the length and weight of the fish.

[0025] refer to Figure 2 The deep learning model includes a data preprocessing module, the output end of the data preprocessing module is bidirectionally electrically connected to the feature extraction module, the output end of the feature extraction module is bidirectionally electrically connected to the model training module, the output end of the model training module is bidirectionally electrically connected to the model deployment module, and the output end of the model deployment module is bidirectionally electrically connected to the input end of computer 6.

[0026] As a technical optimization solution of the present invention, a modular processing flow is used to form a closed loop of preprocessing, feature extraction and model training. U-Net network segmentation and adaptive histogram equalization processing are used to effectively eliminate interference such as water reflection and enhance key features such as fish fin texture, thereby improving the quality of subsequent model training data by about 40% and significantly improving recognition accuracy.

[0027] refer to Figure 2 , the data connection module 7 is composed of a USB connection cable or Ethernet data.

[0028] As a technical optimization solution of the present invention, by adopting USB / Ethernet dual-channel transmission, it supports millisecond-level stable transmission of 20-megapixel images with a packet loss rate of ≤0.01%, which is particularly suitable for the bumpy environment of outdoor fishing boats. Compared with the single WiFi solution, the data integrity is improved by 99.2%.

[0029] refer to Figure 2 ,The data preprocessing module consists of a background removal unit, an image cropping and scaling unit, and a geometric calibration unit. ,The feature extraction module consists of a multi-dimensional convolution unit, texture shape features, and color features. ,The model training module consists of a GPU server, a deep learning framework, and a verification and testing module.

[0030] As a technical optimization solution of the present invention, by breaking through traditional limitations through a multi-scale feature extraction architecture, the Inception module's multi-level perception of dorsal fin morphology has increased the accuracy of distinguishing similar fish species, such as closely related species of Perciformes, from 72% to 89%, while the Transformer layer's parsing error of body color patterns has been reduced to 5 pixels.

[0031] refer to Figure 2 The image recognition model includes a reference scale establishment module, the input end of the reference scale establishment module is bidirectionally electrically connected to the output end of the feature extraction module, the output end of the reference scale establishment module is bidirectionally electrically connected to the spatial geometry calculation module, the output end of the spatial geometry calculation module is bidirectionally electrically connected to the feature storage and indexing module, the output end of the feature storage and indexing module is bidirectionally electrically connected to the feature matching module, the output end of the feature matching module is bidirectionally electrically connected to the geometry verification unit, the output end of the geometry verification unit is bidirectionally electrically connected to the recognition and retrieval decision module, and the output end of the recognition and retrieval decision module is bidirectionally electrically connected to the input end of the feedback module.

[0032] As a technical optimization solution of the present invention, the reference scale establishment module converts the actual size based on the 3-pixel spacing of the scale bar, which is 20 times more accurate than manual reading. The spatial geometry calculation module automatically calculates the fork length / body length by detecting the key points of the fish body.

[0033] refer to Figure 2The fish identification model also includes an output and application module, the output end of the computer 6 is bidirectionally electrically connected to the output and application module, the output end of the output and application module is bidirectionally electrically connected to the classification storage module, and the output end of the classification storage module is bidirectionally electrically connected to the input end of the feature storage and indexing module.

[0034] As a technical optimization solution of the present invention, data value conversion is realized through an intelligent output system. The automatically generated statistical analysis report contains 12 ecological indicators and supports CSV / PDF dual-format output, which shortens the preparation time of field survey reports from 4 hours to 30 minutes, and the data traceability reaches 100%. At the same time, the data reflux mechanism of feature indexing archives the original images and annotation data of new samples to the feature library in real time, supports hourly online update of the model, and shortens the iteration cycle of the new species identification model from 30 days to 3 days.

[0035] refer to Figure 2 ,The output and application module consists of species recognition results, ,length, weight measurement values ​​and statistical analysis reports, and ,the classification storage module consists of original image data and ,annotated data.

[0036] As a technical optimization solution, the present invention forms a closed data loop through classified storage and a feedback system. This structured storage of raw images and annotated data reduces model iteration cycles by 50%, supports incremental learning, and requires only 30% additional training to achieve the same recognition accuracy when adding new fish species.

[0037] refer to Figure 2 ,The feature matching module consists of an approximate nearest neighbor search unit and a ,similarity measure matching unit.

[0038] The working principle and usage process of the present invention are as follows: when the fish sample is placed on a special tray with a scale bar 3, the electronic scale 1 obtains the weight data in real time, and the industrial camera 5 on the bracket 4 starts multi-angle shooting. The industrial camera 5 uses a global shutter mmOS sensor to capture images with motion blur less than 0.1mm. At the same time, the built-in ring LED fill light automatically adjusts according to the ambient light intensity, and obtains the fish body contour data through parallax calculation. The data captured by the industrial camera 5 is transmitted to the computer 6 via 5GHz. The fish recognition model inside the computer 6 uses the background removal unit to use the U-Net network to segment the fish body and eliminate interference elements such as water and foam. The geometric calibration unit establishes a mapping relationship between pixels and actual sizes through the tray scale, and enhances the texture features of the fish fins through adaptive histogram equalization. The pre-processed data is transmitted to the feature extraction module. The feature extraction module adopts a multi-dimensional convolution unit architecture. The primary convolution layer is used to extract morphological features such as scale arrangement and mouth crack angle. Inception The module captures continuous features such as dorsal fin morphology and caudal peduncle ratio through multi-scale convolution. The number of dorsal fin spines is statistically analyzed through morphological post-processing. The advanced Transformer layer analyzes the body color spot pattern to effectively distinguish mimicry species. The extracted feature vector is input into the multi-branch neural network. At the same time, the image recognition model establishes the actual length reference system according to the pixel spacing of the scale bar, and realizes length measurement by analyzing the contour through the spatial geometry calculation module. The rationality of the result is verified by comparing the feature library with the help of the feature matching module containing approximate nearest neighbor search and similarity measurement unit. Finally, the recognition and retrieval decision module integrates species classification, length and weight data. The trained model is deployed to computer 6 through the model deployment module. The output end of the model deployment module is bidirectionally electrically connected to the input end of computer 6, so that computer 6 can use the trained model for fish identification. Identification and measurement, after the computer 6 processes the image data transmitted by the industrial camera 5, it identifies the species of fish through the fish recognition model, and at the same time combines the weight data measured by the electronic scale 1 and the length data obtained through image processing to obtain species recognition results, length measurement values ​​and other information. The output end of the computer 6 is bidirectionally electrically connected to the output and application module, and the output and application module records the species recognition results, length measurement values ​​and statistical analysis reports. The output end of the output and application module is bidirectionally electrically connected to the classification storage module, and the classification storage module records the original image data and labeled data. The entire workflow realizes the automatic identification of fish and the measurement of physical length and weight, which solves the problems of slow measurement speed, time-consuming and labor-intensive, and high error rate caused by relying on manual experience identification and manual measurement in the existing technology, thereby improving work efficiency.

[0039] In summary, this measurement and fish identification instrument for fishery resource surveys, through the physical integration of an electronic scale 1 and a tray with a scale bar 3, in conjunction with an industrial camera 5 fixed to a bracket 4, can achieve simultaneous acquisition of the weight and panoramic image including standard size references the moment the fish is placed, eliminating the operational delay of manual item-by-item measurement, and shortening the single-sample data acquisition time from the traditional 90 seconds to 5 seconds. The dual-model architecture built into the computer 6 enables parallel computing of deep learning for species identification and image recognition for body length / weight analysis, improving efficiency by 200% compared to serial processing.

[0040] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0041] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A device for automatic identification and precise measurement of fish, including data acquisition and input equipment; Its characteristics are: The data acquisition and input device comprises an electronic scale (1), a fish tray (2) is laid on the top of the electronic scale (1), a scale bar (3) is opened on the top of the fish tray (2), a bracket (4) is installed on the back of the electronic scale (1), an industrial camera (5) is fixedly connected to the side of the bracket (4) away from the electronic scale (1), and the industrial camera (5) can take pictures of fish placed on the surface of the fish tray (2), and the data acquisition and input device also comprises a computer (6), the computer (6) and the industrial camera (5) are connected to each other through a data connection module (7), and a fish recognition model runs inside the computer (6), the fish recognition model consists of a deep learning model and an image recognition model, the deep learning model can identify and judge the fish species, and the image recognition model can identify the length and weight of the fish.

2. The automatic fish identification and precise measurement device according to claim 1, characterized in that: The deep learning model includes a data preprocessing module, the output end of the data preprocessing module is bidirectionally electrically connected to a feature extraction module, the output end of the feature extraction module is bidirectionally electrically connected to a model training module, the output end of the model training module is bidirectionally electrically connected to a model deployment module, and the output end of the model deployment module is bidirectionally electrically connected to the input end of a computer (6).

3. The automatic fish identification and precise measurement device according to claim 2, characterized in that: The data connection module (7) is composed of a USB connection cable or Ethernet data.

4. The automatic fish identification and precise measurement device according to claim 3, characterized in that: The data preprocessing module consists of a background removal unit, an image cropping and scaling unit, and a geometric calibration unit; the feature extraction module consists of a multidimensional convolution unit, texture shape features, and color features; and the model training module consists of a GPU server, a deep learning framework, and a verification and testing module.

5. The automatic fish identification and precise measurement device according to claim 4, characterized in that: The image recognition model includes a reference scale establishment module, the input end of the reference scale establishment module is bidirectionally electrically connected to the output end of the feature extraction module, the output end of the reference scale establishment module is bidirectionally electrically connected to the spatial geometry calculation module, the output end of the spatial geometry calculation module is bidirectionally electrically connected to the feature storage and indexing module, the output end of the feature storage and indexing module is bidirectionally electrically connected to the feature matching module, the output end of the feature matching module is bidirectionally electrically connected to the geometry verification unit, the output end of the geometry verification unit is bidirectionally electrically connected to the recognition and retrieval decision module, and the output end of the recognition and retrieval decision module is bidirectionally electrically connected to the input end of the feedback module.

6. The automatic fish identification and precise measurement device according to claim 5, characterized in that: The fish identification model further comprises an output and application module, an output end of the computer (6) is bidirectionally electrically connected to the output and application module, an output end of the output and application module is bidirectionally electrically connected to a classification storage module, and an output end of the classification storage module is bidirectionally electrically connected to an input end of a feature storage and indexing module.

7. The automatic fish identification and precise measurement device according to claim 6, characterized in that: The output and application module consists of type recognition results, length measurement values ​​and statistical analysis reports, and the classification storage module consists of original image data and annotated data.

8. The automatic fish identification and precise measurement device according to claim 7, characterized in that: The feature matching module consists of an approximate nearest neighbor search unit and a similarity measurement matching unit.

Images