A fish sorting method based on a color sorter

Abstract

The present invention discloses a fish sorting method based on a color sorter, which relates to the technical field of fish classification by color sorters. First, a number of pictures of all fish to be classified are collected; each picture is classified and marked; individual entities of experimental fish are collected and placed into the color sorter, and the fish are separated by color through the color sorter; all separated fish groups are obtained from the separation results; for each fish group, a neural network model for identifying the fish species in the fish group is trained with the pictures in the corresponding picture set as inputs; the trained neural network model is installed in the color sorter, and during the actual sorting process, the fish species are analyzed by combining the color sorter and the trained neural network model; thus, the problems of insufficient color accuracy of the color sorter and excessive computing power required for the neural network model to identify are avoided.

Description

Technical Field

[0001] The present invention belongs to the field of color sorters, relates to fish classification technology, and specifically is a fish sorting method based on a color sorter. Background Art

[0002] A color sorter is a device that automatically sorts different colored items based on the differences in the optical characteristics of fish using photoelectric detection technology; large color sorters are used in the fields of fish detection and classification; due to the color accuracy problem of color sorters, for different varieties of fish, if only distinguished by color, errors often occur; and using a neural network model to identify all fish often causes problems of excessive time and computing power for training the neural network.

[0003] Therefore, a fish sorting method based on a color sorter is proposed. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art. For this purpose, the present invention proposes a fish sorting method based on a color sorter. This fish sorting method based on a color sorter combines the color sorter with a neural network model. After the fish are first grouped by the color sorter once, then the neural network model is used to identify each group of fish, avoiding the problems of insufficient color accuracy of the color sorter and excessive computing power required for the neural network model to identify.

[0005] To achieve the above object, according to an embodiment of the first aspect of the present invention, a fish sorting method based on a color sorter is proposed, including the following steps:

[0006] Step 1: Collect a number of pictures of all fish to be classified in advance; classify the fish pictures by species; and mark each picture.

[0007] Step 2: Collect all individual entities of the fish to be sorted, and put all the fish into the color sorter, and separate the fish by color through the color sorter.

[0008] Step 3: Obtain all the separated fish groups from the separation result; it should be understood that the colors of the fish in each group separated by the color sorter are the same or similar; mark each group of fish as P; then mark each fish variety in group P as Pf.

[0009] Step 4: For each fish variety Pf in the fish group P, select the image corresponding to the fish variety Pf from the collected pictures; obtain the picture set of all fish varieties in the fish group P.

[0010] Step 5: For fish group P, use the pictures in the corresponding picture set as input to train a neural network model for identifying the fish species in fish group P; that is, each neural network model will correspond to a sorted color.

[0011] Step 6: Install the trained neural network model in the color sorter. During the actual sorting process, analyze the fish species by combining the color sorter and the trained neural network model.

[0012] Step 7: Display the sorting results on the human-machine interaction interface of the color sorter.

[0013] The way to mark fish pictures is to use integers to mark the pictures of each type of fish; as an example: grass carp is marked as 0, black carp is marked as 1, etc.

[0014] The number of individual entities of all the collected fish is obtained through prior experiments based on actual experience; ensure that the sorting result of the color sorter is to sort fish with the same or similar colors into the same group. It should be understood that the individual entities of this part of the fish are used for the color grouping experiment of the fish by the color sorter. Therefore, all the fish species to be classified need to be included.

[0015] The color sorter sorts fish through the following steps:

[0016] Step P1: The fish enter the machine from the top hopper, and the selected fish fall along the feeding and distributing trough under the vibration of the feeding device.

[0017] Step P2: The fish pass through the vibrator at the upper end of the chute and slide into the sorting box along the chute at an accelerated speed.

[0018] Step P3: After entering the sorting box, they pass through between the image processing sensor CCD and the background device; under the action of the light source, the CCD receives the synthetic light signal from the selected fish, causing the system to generate an output signal, which is amplified and then transmitted to the FPGA+ARM computing and processing system, and then the control system issues an instruction to drive the jet solenoid valve to blow different-colored particles into the waste cavity of the hopper and flow away.

[0019] Step P4: The selected fish continue to fall into the finished product cavity of the hopper and flow out, so that the selected fish achieve the purpose of being carefully selected.

[0020] Training the neural network model for identifying the fish species in fish group P includes the following steps:

[0021] Step Q1: For each fish group P, collect pictures of each fish breed in this group, and preprocess the pixel size of each picture to the size acceptable by the neural network model to form the input set of the neural network model; preferably, the neural network model is a CNN neural network; the preprocessing method can be scaling or zero-padding.

[0022] Step Q2: For each fish group, use each group of picture sets as input and input them into the neural network model for training.

[0023] Step Q3: Set the recognition accuracy threshold K in advance according to actual experience; during the training process of the neural network model, when the recognition accuracy reaches the recognition accuracy threshold K, stop the training.

[0024] The process of analyzing fish breeds includes the following steps:

[0025] Step X1: Sort all the fish to be recognized through a color sorter; and from all the neural network models, find the trained neural network model corresponding to the color of each group of fish.

[0026] Step X2: For each fish individual in each group of fish, use an image capture device to obtain its image.

[0027] Step X3: Use this image as input and input it into the neural network model corresponding to the fish group where this fish is located.

[0028] Step X4: Obtain the output of each fish image in the neural network model.

[0029] The way to display the recognition result on the human-machine interaction interface of the color sorter is: for each fish, when using the neural network model to output the breed of this fish, the human-machine interaction interface displays the image of this fish and the output breed.

[0030] Compared with the prior art, the beneficial effects of the present invention are:

[0031] The present invention combines the color sorter with the neural network model. First, the fish are grouped once through the color sorter, and then the neural network model is used to identify each group of fish, avoiding the problems of insufficient color accuracy of the color sorter and excessive computing power required for the neural network model to identify. Brief Description of the Drawings

[0032] Figure 1 It is a flowchart of the present invention. Detailed Embodiment

[0033] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative work shall fall within the protection scope of the present invention.

[0034] As Figure 1 shown, a fish sorting method based on a color sorter includes the following steps:

[0035] Step 1: Collect a number of pictures of all fish to be classified in advance; classify the fish pictures by species; and mark each picture.

[0036] Step 2: Collect the individual entities of all fish to be sorted, and put all the fish into the color sorter, and separate the fish by color through the color sorter.

[0037] Step 3: Obtain all the separated fish groups from the separation results; it should be understood that the colors of the fish in each group separated by the color sorter are the same or similar; mark each group of fish as P; then each fish species in group P is marked as Pf.

[0038] Step 4: For each fish species Pf in the fish group P, select the image corresponding to the fish species Pf from the collected pictures; obtain the picture set of all fish species in the fish group P.

[0039] Step 5: For the fish group P, use the pictures in the corresponding picture set as the input to train a neural network model for identifying the fish species in the fish group P; that is, each neural network model will correspond to a sorted color.

[0040] Step 6: Install the trained neural network model in the color sorter, and in the actual sorting process, analyze the fish species by combining the color sorter and the trained neural network model.

[0041] Step 7: Display the sorting results on the man-machine interaction interface of the color sorter.

[0042] A color sorter is a device that automatically sorts different-colored items using photoelectric detection technology based on the differences in the optical characteristics of fish; the color sorter is used in the fields of bulk fish or packaging industrial products and food quality inspection and grading; due to the problem of color accuracy of the color sorter, for different species of fish, if only distinguished by color, errors often occur.

[0043] In this embodiment, the method of marking the fish pictures is to use integers as marks for the pictures of each type of fish; as an example: grass carp is marked as 0, black carp is marked as 1, etc.

[0044] In this embodiment, the number of individual entities of all the fish collected is obtained in advance through actual empirical experiments; it is ensured that the sorting result of the color sorter is to sort the fish with the same or similar colors into the same group; it should be understood that the individual entities of this part of the fish are used for the color grouping experiment of the fish by the color sorter, so all the fish varieties to be classified need to be included;

[0045] In this embodiment, the sorting of fish by the color sorter includes the following steps:

[0046] Step P1: The fish enter the machine from the top aggregate hopper, and the selected fish fall along the feeding and distributing chute under the vibration of the feeding device;

[0047] Step P2: The fish pass through the vibrator at the upper end of the chute and slide into the sorting box along the chute at an accelerating speed;

[0048] Step P3: After entering the sorting box, they pass between the image processing sensor CCD and the background device; under the action of the light source, the CCD receives the synthetic light signal from the selected fish, causing the system to generate an output signal, which is amplified and then transmitted to the FPGA+ARM computing and processing system, and then the control system issues an instruction to drive the injection solenoid valve to blow the particles of different colors into the waste cavity of the feed hopper and flow away;

[0049] Step P4: The selected fish continue to fall into the finished product cavity of the hopper and flow out, so that the selected fish achieve the purpose of being carefully selected;

[0050] In this embodiment, training the neural network model for identifying the fish varieties in the fish grouping P includes the following steps:

[0051] Step Q1: For each fish grouping P, collect the pictures of each fish variety in this grouping, and preprocess the pixel size of each picture to the size acceptable by the neural network model to form the input set of the neural network model; preferably, the neural network model is a CNN neural network; the preprocessing method can be in the way of scaling or padding with zeros;

[0052] Step Q2: For each fish grouping, use each group of picture sets as the input and input them into the neural network model for training; preferably, the neural network model outputs the numerical labels of each fish, with the actual numerical labels of the fish as the recognition target; the recognition accuracy rate of the numerical labels output by the neural network model for the actual numerical labels is the training target;

[0053] Step Q3: Set the recognition accuracy threshold K in advance according to actual experience; during the training process of the neural network model, when the recognition accuracy reaches the recognition accuracy threshold K, stop the training; it should be understood that the training process of the neural network model includes setting and adjusting the training parameters of the neural network model according to actual experience;

[0054] In this embodiment, the process of analyzing fish species includes the following steps:

[0055] Step X1: Sort all the fish to be recognized through a color sorter; and from all the neural network models, find the trained neural network model corresponding to the color of each group of fish;

[0056] Step X2: For each fish individual in each group of fish, use an image capture device to obtain its image;

[0057] Step X3: Take this image as the input and input it into the neural network model corresponding to the fish grouping where the fish is located;

[0058] Step X4: Obtain the output of each fish image in the neural network model; this output is the variety of each fish recognized by the neural network model;

[0059] In this embodiment, the way to display the recognition result on the man-machine interaction interface of the color sorter is: for each fish, when using the neural network model to output the variety of this fish, the man-machine interaction interface displays the image of this fish and the output variety; preferably, the color sorter can selectively release each fish into the finished product cavity in turn.

[0060] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A fish sorting method based on a color sorter, characterized in that, it includes the following steps: Step 1: Collect a number of pictures of all fish to be classified in advance; classify the fish pictures by species; and mark each picture; Step 2: Collect all individual entities of the fish to be sorted, and put all the fish into the color sorter, and separate the fish by color through the color sorter; Step 3: Obtain all the separated fish groups from the separation results; mark each group of fish as P; then each fish variety in group P is marked as Pf; Step 4: For each fish variety Pf in each fish group P, select the image corresponding to the fish variety Pf from the collected pictures; Obtain the picture set of all fish varieties in fish group P; Step 5: For fish group P, use the pictures in the corresponding picture set as input to train a neural network model for identifying the fish varieties in fish group P; Step 6: Install the trained neural network model in the color sorter. During the actual sorting process, analyze the fish varieties by combining the color sorter and the trained neural network model; Step 7: Display the sorting results on the man-machine interaction interface of the color sorter; Training the neural network model for identifying the fish varieties in fish group P includes the following steps: Step Q1: For each fish group P, collect the pictures of each fish variety in this group, and preprocess the pixel size of each picture to the size acceptable by the neural network model to form the input set of the neural network model; Step Q2: For each fish group, use each group of picture sets as input and input them into the neural network model for training; Step Q3: Set the recognition accuracy threshold K in advance according to actual experience; when the neural network model is training, when the recognition accuracy reaches the recognition accuracy threshold K, stop training.

2. The fish sorting method based on a color sorter according to claim 1, characterized in that, The way of marking the fish pictures is to use integers as marks for the pictures of each type of fish.

3. The fish sorting method based on a color sorter according to claim 1, characterized in that, The preprocessing method is the method of scaling or padding with zeros.

4. The fish sorting method based on a color sorter according to claim 1, characterized in that, The process of analyzing the fish varieties includes the following steps: Step X1: Sort all the fish to be recognized through the color sorter; and find the trained neural network model corresponding to the color of each group of fish from all the neural network models; Step X2: For each fish individual in each group of fish, use an image capture device to obtain its image; Step X3: Take this image as input and input it into the neural network model corresponding to the fish group where the fish is located; Step X4: Obtain the output of each fish image in the neural network model.

5. The fish sorting method based on a color sorter according to claim 1, characterized in that, The way to display the recognition result on the human-machine interaction interface of the color sorter is as follows: for each fish, when using the neural network model to output the variety of the fish, the human-machine interaction interface displays the image of the fish and the output variety.

Images