Intelligent fish identification method and system for fishery fishing in small reservoir of hilly pond

Abstract

The invention relates to the technical field of fishery fishing, in particular to an intelligent fish identification method and system for fishery fishing in a hilly pond small reservoir, and the method comprises the following steps: collecting an original image of a fish school in the hilly pond small reservoir in real time; using a multi-strategy fusion enhancement scheme to pre-process the original fish school image to obtain a clear fish school image; constructing a lightweight fish identification model, and using the lightweight fish identification model to identify the clear image of the fish school; and grading the fishes in the clear fish school image according to an identification result, and estimating the proportion of the fishes of different grades in the fish school. According to the method, the accuracy of identifying and grading the fishes in the small hilly pond reservoir can be improved, and a basis is provided for on-demand preferential fishing and improvement of the fishing efficiency of the small hilly pond reservoir.

Description

Technical Field

[0001] The present invention relates to the technical field of fishery, and in particular to a fish intelligent identification method and system for fishery in mountain ponds and small reservoirs. Background Art

[0002] To improve the efficiency and accuracy of fish harvesting, intelligent recognition technology has been gradually introduced into the aquaculture industry in recent years. Using advanced methods such as deep learning algorithms and image recognition, intelligent recognition systems can automatically identify and classify fish in mountain ponds and small reservoirs. However, existing intelligent recognition technology still faces some challenges when applied to fisheries in mountain ponds and small reservoirs.

[0003] On the one hand, the environment of small mountain ponds and reservoirs is complex and changeable. Factors such as lighting conditions, water quality, and fish activity patterns can affect the accuracy and stability of intelligent recognition systems. On the other hand, different species of fish can have significant similarities in appearance, especially when they are at similar growth stages. This can make it difficult for intelligent recognition systems to distinguish between different species.

[0004] Therefore, in order to improve the efficiency and accuracy of fishing operations, it is necessary to explore a more accurate and reliable method for intelligent fish identification. Summary of the Invention

[0005] In view of the defects in the prior art, the present invention provides a fish intelligent identification method and system for fishing in mountain ponds and small reservoirs.

[0006] To achieve the above objectives, in a first aspect, the present invention provides an intelligent fish identification method for use in mountain pond and small reservoir fishing. The method comprises: collecting raw images of fish schools in mountain ponds and small reservoirs in real time; preprocessing the raw images of the fish schools using a multi-strategy fusion enhancement scheme to obtain clear images of the fish schools; constructing a lightweight fish identification model and using the lightweight fish identification model to identify the clear images of the fish schools; and grading the fish in the clear images of the fish schools based on the identification results, and estimating the proportion of fish of different grades within the fish school. The present invention can improve the accuracy of fish identification and grading in mountain ponds and small reservoirs, providing a basis for on-demand and preferential fishing and improving fishing efficiency in mountain ponds and small reservoirs.

[0007] Optionally, the preprocessing of the original image of the school of fish using a multi-strategy fusion enhancement solution to obtain a clear image of the school of fish comprises the following steps: Filtering the original fish school image using homomorphic filtering to obtain a first fish school image; Performing color restoration on the original image of the school of fish using a multi-scale retinal cortex theory algorithm with color restoration to obtain a second image of the school of fish; Dehazing the original image of the school of fish using an image dehazing algorithm based on a dark channel prior to obtain a third image of the school of fish; The first fish school image, the second fish school image and the third fish school image are fused to obtain a clear image of the fish school.

[0008] Optionally, fusing the first school of fish image, the second school of fish image, and the third school of fish image to obtain the clear school of fish image comprises the following steps: Splitting the first fish school image and the second fish school image into RGB three-channels to obtain a first group of monochrome images and a second group of monochrome images, respectively; fusing the monochrome images of the same channel in the first group of monochrome images and the second group of monochrome images according to a monochrome fusion model to obtain a first group of monochrome fused images; Perform RGB three-channel fusion on the first set of monochrome fused images to obtain an intermediate image; Splitting the third fish school image and the intermediate image into RGB three-channels to obtain a third group of monochrome images and a fourth group of monochrome images, respectively; fusing the third group of monochrome images and the monochrome images of the same channel in the fourth group of monochrome images according to a monochrome fusion model to obtain a second group of monochrome fused images; The second group of monochrome fused images is subjected to RGB three-channel fusion to obtain a clear image of the fish school.

[0009] Optionally, the monochrome fusion model satisfies the following relationship:

[0010] in, is the pixel value of the rth pixel in the monochrome fused image; n is the total number of pixels in any one of the two color images involved in the fusion; m is the number of pixels in the neighborhood of any pixel in the two color images involved in the fusion; is the pixel value of the i-th pixel on the k-th color image of the two color images involved in the fusion; is the pixel value of the jth pixel in the neighborhood of the i-th pixel on the k-th color image in the two color images involved in the fusion; is the position weight between the i-th pixel point on the k-th color image and the j-th pixel point in its neighborhood in the two color images involved in the fusion; is the pixel value of the rth pixel on the kth monochrome image of the two monochrome images involved in the fusion.

[0011] Optionally, constructing a lightweight fish recognition model and using the lightweight fish recognition model to recognize the clear image of the fish school includes the following steps: Construct an initial fish recognition model; Construct a fish dataset using fish images; Using the fish data set to train and verify the initial fish recognition model, thereby obtaining the lightweight fish recognition model; The lightweight fish recognition model is used to identify the species and growth stages of the fish in the clear image of the fish school.

[0012] Optionally, the initial fish recognition model includes a feature extraction network, a feature splicing layer, a pooling layer, a fully connected layer and an output layer, the feature extraction network includes a feature extraction backbone network and a branch network, the pooling layer includes four sub-pooling layers, and the branch network includes a first branch network, a second branch network and a third branch network; The construction of the initial fish identification model comprises the following steps: Constructing a four-level feature extraction unit to form the feature extraction backbone network, and simultaneously constructing the first branch network, the second branch network, and the third branch network; Sequentially taking the outputs of the feature extraction backbone network, the first branch network, the second branch network, and the third branch network as inputs of the four sub-pooling layers; The outputs of the four sub-pooling layers are used as the inputs of the feature splicing layer, the output of the feature splicing layer is used as the input of the fully connected layer, and the output of the fully connected layer is used as the output layer to obtain an initial fish recognition model.

[0013] Optionally, the feature extraction unit satisfies the following relationship:

[0014] Wherein, F is the feature extracted by the feature extraction unit; P is the clear image of the fish school; Indicates that the convolution kernel is 1×1; and The first sub-feature and the second sub-feature are obtained after performing channel segmentation based on the input features of the feature extraction unit; Indicates that the convolution kernel is 3×3 and the expansion rate is Depthwise separable convolution, b=1,2,3,4; Indicates channel shuffling.

[0015] Optionally, the branch network satisfies the following relationship:

[0016] in, is the feature extracted by the t-th branch network; Indicates that the convolution kernel is 1×1; is the fusion function, The features extracted by the feature extraction unit at level c, c = 1, 2, 3; Indicates maximum pooling; Indicates that the convolution kernel is 3×3 and the expansion rate is Depthwise separable convolution, d=1,2,3; Indicates upsampling; t=c.

[0017] Optionally, grading the fish in the clear image of the fish school according to the recognition result and estimating the proportion of fish of different grades in the fish school comprises the following steps: Based on the identification results, the fish are divided into different grades according to the growth stages of the fish; The average number of fish of different grades in the plurality of clear images of the fish school is calculated, and then the proportion of fish of different grades in the fish school is estimated.

[0018] In a second aspect, the present invention also provides another intelligent fish identification system for mountain pond and small reservoir fishing. The system comprises a data acquisition device, a data output device, a processor, and a memory. The memory comprises a computer-readable storage medium storing a computer program. The computer program comprises program instructions that, when executed by the processor, cause the processor to implement the intelligent fish identification method for mountain pond and small reservoir fishing provided by the present invention. This system can improve the efficiency of intelligent fish identification in mountain ponds and small reservoirs and enhance the practicality of the method provided by the present invention.

[0019] The present invention has the following beneficial effects:

[0020] This method uses homomorphic filtering to enhance the dark details of the original fish image to obtain the first fish image, uses a multi-scale retinal cortex theory algorithm with color restoration to restore the color of the original fish image to obtain the second fish image, uses an image dehazing algorithm based on dark channel prior to dehaze the original fish image to obtain the third fish image, and fuses the three obtained fish images based on a monochrome fusion model to obtain a clear fish image. This solves the problem that a single image enhancement algorithm is difficult to handle image restoration in complex water environments, and provides a solid foundation for the accurate identification of underwater fish. Based on the acquisition of clear images of fish schools, this method further constructs a lightweight fish recognition model. This model has a low number of parameters and can obtain both shallow and deep features of the image through feature extraction and feature fusion, thereby achieving accurate fish identification. This improves the accuracy of fish identification and classification in mountain ponds and small reservoirs, providing a basis for on-demand and preferential fishing and improving fishing efficiency in mountain ponds and small reservoirs. The system can improve the efficiency of intelligent identification of fish in mountain ponds and small reservoirs, and can improve the practicality of the method provided by the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without any creative work.

[0022] Figure 1 Schematic diagram of a flow chart of a method for intelligent fish identification in mountain ponds and small reservoirs according to an embodiment of the present invention; Figure 2 A schematic diagram of the network structure of a branch network according to an embodiment of the present invention; Figure 3 The present invention is a schematic diagram of a framework of an intelligent fish identification system for fishing in mountain ponds and small reservoirs according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] Specific embodiments of the present invention will be described in detail below. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the present invention. In the following description, numerous specific details are set forth to provide a thorough understanding of the present invention. However, it will be apparent to one of ordinary skill in the art that these specific details are not necessarily required to practice the present invention. In other instances, well-known circuits, software, or methods are not specifically described to avoid obscuring the present invention.

[0024] Throughout this specification, references to "one embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with the embodiment or example is included in at least one embodiment of the present invention. Therefore, appearances of the phrases "in one embodiment," "in an embodiment," "an example," or "an example" in various places throughout this specification are not necessarily all referring to the same embodiment or example. Furthermore, the particular features, structures, or characteristics may be combined in any suitable combinations and / or subcombinations in one or more embodiments or examples. Furthermore, those skilled in the art will appreciate that the figures provided herein are for illustrative purposes only and are not necessarily drawn to scale.

[0025] It should be noted in advance that, in an optional embodiment, except for independent explanations, the same symbols or letters appearing in all formulas have the same meanings and values.

[0026] In an alternative embodiment, see Figure 1 The present invention provides a method for intelligently identifying fish for fishing in mountain ponds and small reservoirs, the method comprising the following steps: S1. Real-time collection of original images of fish schools in mountain ponds and small reservoirs.

[0027] Specifically, in this embodiment, underwater fish school images of a mountain pond or small reservoir are collected in real time by an underwater camera, and the obtained underwater fish school images are cropped into images with a height and width of 224 pixels to obtain the original images of the fish school.

[0028] Furthermore, in other optional embodiments, the underwater fish school image may be cropped into images of other heights and widths.

[0029] S2. Use a multi-strategy fusion enhancement solution to preprocess the original image of the fish school to obtain a clear image of the fish school.

[0030] The underwater environment of small mountain ponds and reservoirs is complex, and the original images of fish schools obtained often have problems such as unclear dark details, color cast, and water mist. A single image enhancement algorithm is difficult to cope with the problem of image restoration in complex water environments, which makes it difficult to accurately identify fish based on the original images of fish schools obtained in real time. Therefore, this embodiment considers using a multi-strategy fusion enhancement solution to solve the problems of the original images of fish schools, improve image quality, and thus provide a basis for accurate identification of underwater fish. Step S2 specifically includes the following steps: S21. Filter the original fish school image using homomorphic filtering to obtain a first fish school image.

[0031] Specifically, in this embodiment, using homomorphic filtering to filter the original image of the school of fish can enhance the dark features of the original image of the school of fish and improve the image contrast. However, homomorphic filtering cannot solve the problems of color cast and water mist in the original image of the school of fish.

[0032] S22. Perform color restoration on the original image of the school of fish using a multi-scale retinal cortex theory algorithm with color restoration to obtain a second image of the school of fish.

[0033] Specifically, in this embodiment, the Multi-Scale Retinal Cortex Theory (MSRCR) algorithm with color restoration can resolve the color cast problem in the original fish school image, restoring the original color of the fish school image. However, the MSRCR algorithm cannot resolve the water mist problem in the original fish school image.

[0034] S23. Defogging the original image of the school of fish using an image defogging algorithm based on dark channel prior to obtain a third image of the school of fish.

[0035] Specifically, in this embodiment, the use of an image defogging algorithm based on a dark channel prior can solve the problems of image blur and contrast reduction caused by fog, but it also causes artifacts and color distortion.

[0036] Furthermore, dehazing the original image of the school of fish using an image dehazing algorithm based on a dark channel prior can be achieved through existing technical means, which will not be described in detail here.

[0037] S24: Fusing the first fish school image, the second fish school image, and the third fish school image to obtain a clear image of the fish school.

[0038] As can be seen from the description of steps S21 to S23, a single image enhancement algorithm is difficult to address the problem of image restoration in complex water environments. However, these algorithms each have their own advantages. Therefore, this embodiment fuses the first, second, and third fish school images based on a monochrome fusion model to achieve complementary advantages of different image enhancement algorithms. Step S24 specifically includes the following steps: S241 , performing RGB three-channel splitting on the first school of fish image and the second school of fish image to obtain a first group of monochrome images and a second group of monochrome images, respectively.

[0039] Specifically, in this embodiment, the first group of monochrome images and the second group of monochrome images each include three monochrome images, which are from three channels: R channel, G channel, and B channel.

[0040] Furthermore, the RGB three-channel splitting of the first fish school image and the second fish school image can be achieved by existing technical means, which will not be described in detail here.

[0041] S242. Fuse the monochrome images of the same channel in the first group of monochrome images and the second group of monochrome images according to a monochrome fusion model to obtain a first group of monochrome fused images.

[0042] Specifically, in this embodiment, the monochrome fusion model satisfies the following relationship:

[0043] in, is the pixel value of the rth pixel in the monochrome fused image; n is the total number of pixels in any one of the two color images involved in the fusion; m is the number of pixels in the neighborhood of any pixel in the two color images involved in the fusion; is the pixel value of the i-th pixel on the k-th color image of the two color images involved in the fusion; is the pixel value of the jth pixel in the neighborhood of the i-th pixel on the k-th color image in the two color images involved in the fusion; is the position weight between the i-th pixel point on the k-th color image and the j-th pixel point in its neighborhood in the two color images involved in the fusion; is the pixel value of the rth pixel on the kth monochrome image of the two monochrome images involved in the fusion.

[0044] More specifically, the neighborhood of a pixel is a square area centered on the pixel, specifically including the 8 pixels adjacent to the pixel, so the value of m is 8. For the position weight, if the pixels in the neighborhood of a pixel are located on the diagonal of the neighborhood, the position weight is , otherwise 1.

[0045] It should be noted that the two color images involved in the fusion mentioned in this step are the first fish school image and the second fish school image, and the two monochrome images involved in the fusion are the monochrome images from the same channel in the first and second sets of monochrome images. Specifically, the monochrome image from the R channel in the first and second sets of monochrome images is fused to obtain a monochrome fused image of the R channel. Similarly, monochrome fused images of the G channel and the B channel can be obtained. The pixel value of each pixel in the monochrome fused image is calculated using the monochrome fusion model.

[0046] S243 , performing RGB three-channel fusion on the first group of monochrome fused images to obtain an intermediate image.

[0047] Specifically, in this embodiment, the first group of monochrome fused images includes the monochrome fused image of the R channel, the monochrome fused image of the G channel, and the monochrome fused image of the B channel described in step S242. The process of performing RGB three-channel fusion on these three monochrome fused images to obtain an intermediate image is an existing technology.

[0048] It should be noted that in other optional embodiments, this step can be omitted and the first set of monochrome fused images can be directly used for subsequent processing. In this case, the subsequent step S244 does not need to split the intermediate image into RGB channels, thereby simplifying the process. However, if this step is omitted, some information related to the RGB channels may be lost. By re-fusing the image into RGB format, this information can be retained, thereby achieving better results in subsequent processing.

[0049] S244: Split the third fish school image and the intermediate image into RGB three-channels to obtain a third group of monochrome images and a fourth group of monochrome images, respectively.

[0050] S245 . Fusing the third group of monochrome images and the monochrome images of the same channel in the fourth group of monochrome images according to a monochrome fusion model to obtain a second group of monochrome fused images.

[0051] Specifically, in this embodiment, this step may refer to the content described in step S242.

[0052] S246. Perform RGB three-channel fusion on the second group of monochrome fused images to obtain a clear image of the fish school.

[0053] S3. Construct a lightweight fish recognition model, and use the lightweight fish recognition model to identify the clear image of the fish school.

[0054] Wherein, step S3 specifically includes the following steps: S31. Construct an initial fish recognition model.

[0055] The initial fish recognition model includes a feature extraction network, a feature concatenation layer, a pooling layer, a fully connected layer, and an output layer. The feature extraction network includes a feature extraction backbone network and a branch network. The pooling layer includes four sub-pooling layers. The branch network includes a first branch network, a second branch network, and a third branch network. Step S31 specifically includes the following steps: S311 , constructing a four-level feature extraction unit to form the feature extraction backbone network, and simultaneously constructing the first branch network, the second branch network, and the third branch network.

[0056] Specifically, in this embodiment, the feature extraction unit and the branch network respectively satisfy the following relationships:

[0057]

[0058] Wherein, F is the feature extracted by the feature extraction unit; P is the clear image of the fish school; Indicates that the convolution kernel is 1×1; and The first sub-feature and the second sub-feature are obtained after performing channel segmentation based on the input features of the feature extraction unit; Indicates that the convolution kernel is 3×3 and the expansion rate is Depthwise separable convolution, b=1,2,3,4; Indicates channel shuffling; is the feature extracted by the t-th branch network; is the fusion function, The features extracted by the feature extraction unit at level c, c = 1, 2, 3; Indicates maximum pooling; Indicates that the convolution kernel is 3×3 and the expansion rate is Depthwise separable convolution, d=1,2,3; Indicates upsampling; t=c. Since the relationship between the branch network is relatively complex, for ease of understanding, the network structure of the branch network can be seen in Figure 2 It should be noted that For example, it actually represents the result of convolution of a certain data with a convolution kernel of 1×1. Without affecting the understanding, this embodiment is simplified. In short, it means "convolution with a convolution kernel of 1×1", and the same applies to other similar symbols.

[0059] Further, and Specifically, the input features of the feature extraction unit are subjected to a depthwise separable convolution with a convolution kernel of 3×3 and a dilation rate of 1, and then obtained by channel segmentation. The input features of the first-level feature extraction unit are clear images of fish schools, and the input features of the second to fourth-level feature extraction units are all features extracted by the feature extraction unit of the previous level. The dilation rate combination of the first and third-level feature extraction units is: All are (2, 4, 6, 8), the expansion rate combination of the second and fourth level feature extraction units The expansion rate combinations used in the first to third branch networks are (1, 3, 5, 7). The pooling window size used for the maximum pooling in the branch network is 2×2, and the upsampling in the branch network is implemented by bilinear interpolation.

[0060] Different types of fish may have great similarities in appearance features, especially when they are at similar growth stages. Therefore, the feature extraction backbone network of this embodiment adopts multi-level feature extraction units to fully extract image features step by step to improve the accuracy of fish identification. On the basis of the feature extraction backbone network, this embodiment uses multiple branch networks to fuse the input features of the previous level feature extraction unit with the input features of the next level feature extraction unit, and further extract features to avoid the loss of detailed features that may be caused when the feature extraction unit extracts image features step by step, thereby further improving the accuracy of fish identification. In addition, the feature extraction unit and the branch network structure are relatively simple, and the parameters involved are also relatively few, which can reduce the requirements of the lightweight fish identification model for hardware equipment, facilitate maintenance, and enhance the practicality and applicability of the lightweight fish identification model.

[0061] In other optional embodiments, the convolution kernel size, dilation rate combination, and pooling window size mentioned in this step can be adjusted and optimized according to actual needs.

[0062] S312. Sequentially use the outputs of the feature extraction backbone network, the first branch network, the second branch network, and the third branch network as inputs of the four sub-pooling layers.

[0063] Specifically, in this embodiment, the four sub-pooling layers all perform average pooling on the corresponding inputs, and the pooling window size selected for average pooling is 2×2. The four sub-pooling layers contained in the pooling layer are sequentially recorded as the first sub-pooling layer, the second sub-pooling layer, the third sub-pooling layer, and the fourth sub-pooling layer. The output of the feature extraction backbone network is used as the input of the first sub-pooling layer, the output of the first branch network is used as the input of the second sub-pooling layer, the output of the second branch network is used as the input of the third sub-pooling layer, and the output of the third branch network is used as the input of the fourth sub-pooling layer.

[0064] S313. Use the outputs of the four sub-pooling layers as the inputs of the feature splicing layer, use the output of the feature splicing layer as the input of the fully connected layer, and use the output of the fully connected layer as the output layer to obtain an initial fish recognition model.

[0065] Specifically, in this embodiment, the output of the four sub-pooling layers is used as the input of the feature splicing layer, and then the feature splicing layer crops all the input feature maps into 7×7 feature maps and performs feature splicing, and the result of feature splicing is input into the fully connected layer, and the output of the fully connected layer is used as the output layer, and finally the initial fish recognition model is obtained.

[0066] Furthermore, the operation of feature splicing and the construction of the fully connected layer can refer to the existing technology and will not be described in detail here.

[0067] S32. Use fish images to build a fish dataset.

[0068] Specifically, in this embodiment, images of different species of fish at different growth stages are collected from mountain ponds and small reservoirs to construct a fish dataset.

[0069] S33. Using the fish data set to train and verify the initial fish recognition model, thereby obtaining the lightweight fish recognition model.

[0070] Specifically, in this embodiment, the fish data set is divided into a training set and a validation set in a ratio of 8:2, and then the training set and the validation set are used to complete the training and verification of the initial fish recognition model, and finally a lightweight fish recognition model is obtained.

[0071] S34. Using the lightweight fish recognition model, identify the species and growth stage of the fish in the clear image of the school of fish.

[0072] Specifically, in this embodiment, by inputting a clear image of a school of fish into a lightweight fish recognition model, the species and growth stage of the fish can be identified.

[0073] S4. Classify the fish in the clear image of the fish school according to the recognition result, and estimate the proportion of fish of different grades in the fish school.

[0074] Wherein, step S4 specifically includes the following steps: S41. Based on the recognition result, the fish are divided into different grades according to the growth stages of the fish.

[0075] Specifically, in this embodiment, excluding the embryonic stage of fish, the growth stages of fish can be roughly divided into four growth stages: larval stage, fry stage, young fish stage and adult stage. Therefore, the grades of fish can be divided into grades 1 to 4 accordingly.

[0076] S42. Calculate the average number of fish of different grades in the plurality of clear images of the fish school, and then estimate the proportion of fish of different grades in the fish school.

[0077] Specifically, in this embodiment, the average number of fish of different grades in a plurality of clear images of a school of fish is calculated, and then the proportion of fish of different grades in the school of fish can be calculated.

[0078] It should be noted that, in some cases, the actions described in the specification can be performed in a different order and still achieve the desired results. In this embodiment, the order of steps given is only to make the embodiment appear clearer and easier to explain, rather than to limit it.

[0079] In an alternative embodiment, see Figure 2 The present invention also provides a fish intelligent identification system for mountain pond and small reservoir fishing. The fish intelligent identification system for mountain pond and small reservoir fishing includes: a data acquisition device 1, a data output device 2, a processor 3 and a storage 4. The storage 4 includes a computer-readable storage medium. The computer-readable storage medium stores a computer program. The computer program includes program instructions. When the program instructions are executed by the processor 3, the processor 3 implements a fish intelligent identification method for mountain pond and small reservoir fishing of this embodiment, specifically executing the contents described in steps S1 to S4.

[0080] Specifically, in this embodiment, the data acquisition device 1 includes an underwater camera that can collect original images of fish schools in mountain ponds and small reservoirs in real time, and the data output device 2 includes a digital display screen that can display the recognition results of clear images of fish schools.

[0081] Furthermore, since the system operates based on the method, the system can improve the efficiency of intelligent identification of fish in mountain ponds and small reservoirs, and can improve the practicality of the method provided by the present invention.

[0082] In summary, the method enhances the dark details of the original image of the fish school through homomorphic filtering to obtain a first fish school image, uses a multi-scale retinal cortex theory algorithm with color restoration to restore the color of the original image of the fish school to obtain a second fish school image, uses an image dehazing algorithm based on dark channel prior to dehaze the original image of the fish school to obtain a third fish school image, and fuses the three obtained fish school images based on a monochrome fusion model to obtain a clear image of the fish school, which solves the problem that a single image enhancement algorithm is difficult to cope with image restoration in a complex water environment, and provides a solid foundation for the accurate identification of underwater fish; on the basis of obtaining a clear image of the fish school, the method further constructs a lightweight fish recognition model, which has a low number of parameters and can obtain shallow and deep features of the image through feature extraction and feature fusion, thereby realizing accurate identification of fish, improving the accuracy of fish identification and grading in mountain ponds and small reservoirs, and providing a basis for on-demand preferential fishing and improving the fishing efficiency of mountain ponds and small reservoirs; the system can improve the efficiency of intelligent identification of fish in mountain ponds and small reservoirs, and can improve the practicality of the method provided by the present invention.

[0083] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in this field should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered by the scope of the claims and description of the present invention.

Claims

1. A fish intelligent identification method for fishing in mountain ponds and small reservoirs, characterized in that: The steps include: Real-time collection of original images of fish schools in mountain ponds and small reservoirs; Preprocessing the original image of the school of fish using a multi-strategy fusion enhancement solution to obtain a clear image of the school of fish; Constructing a lightweight fish recognition model, and using the lightweight fish recognition model to identify the clear image of the fish school; The fish in the clear image of the fish school are graded according to the recognition result, and the proportions of fish of different grades in the fish school are estimated.

2. The intelligent fish identification method for fishing in mountain ponds and small reservoirs according to claim 1 is characterized in that: The method of pre-processing the original image of the school of fish using the multi-strategy fusion enhancement scheme to obtain a clear image of the school of fish comprises the following steps: Filtering the original fish school image using homomorphic filtering to obtain a first fish school image; Performing color restoration on the original image of the school of fish using a multi-scale retinal cortex theory algorithm with color restoration to obtain a second image of the school of fish; Dehazing the original image of the school of fish using an image dehazing algorithm based on a dark channel prior to obtain a third image of the school of fish; The first fish school image, the second fish school image and the third fish school image are fused to obtain a clear image of the fish school.

3. The intelligent fish identification method for fishing in mountain ponds and small reservoirs according to claim 2 is characterized in that: The fusing of the first fish school image, the second fish school image, and the third fish school image to obtain the clear fish school image comprises the following steps: Splitting the first fish school image and the second fish school image into RGB three-channels to obtain a first group of monochrome images and a second group of monochrome images, respectively; fusing the monochrome images of the same channel in the first group of monochrome images and the second group of monochrome images according to a monochrome fusion model to obtain a first group of monochrome fused images; Perform RGB three-channel fusion on the first set of monochrome fused images to obtain an intermediate image; Splitting the third fish school image and the intermediate image into RGB three-channels to obtain a third group of monochrome images and a fourth group of monochrome images, respectively; fusing the third group of monochrome images and the monochrome images of the same channel in the fourth group of monochrome images according to a monochrome fusion model to obtain a second group of monochrome fused images; The second group of monochrome fused images is subjected to RGB three-channel fusion to obtain a clear image of the fish school.

4. The intelligent fish identification method for fishing in mountain ponds and small reservoirs according to claim 3 is characterized in that: The monochrome fusion model satisfies the following relationship: ， in, is the pixel value of the rth pixel in the monochrome fused image; n is the total number of pixels in any one of the two color images involved in the fusion; m is the number of pixels in the neighborhood of any pixel in the two color images involved in the fusion; is the pixel value of the i-th pixel on the k-th color image of the two color images involved in the fusion; is the pixel value of the jth pixel in the neighborhood of the i-th pixel on the k-th color image in the two color images involved in the fusion; is the position weight between the i-th pixel point on the k-th color image and the j-th pixel point in its neighborhood in the two color images involved in the fusion; is the pixel value of the rth pixel on the kth monochrome image of the two monochrome images involved in the fusion.

5. The intelligent fish identification method for mountain pond and small reservoir fishing according to claim 1 is characterized in that: Constructing a lightweight fish recognition model and using the lightweight fish recognition model to identify the clear image of the fish school includes the following steps: Construct an initial fish recognition model; Construct a fish dataset using fish images; Using the fish data set to train and verify the initial fish recognition model, thereby obtaining the lightweight fish recognition model; The lightweight fish recognition model is used to identify the species and growth stages of the fish in the clear image of the fish school.

6. The intelligent fish identification method for fishing in mountain ponds and small reservoirs according to claim 5, characterized in that: The initial fish recognition model includes a feature extraction network, a feature splicing layer, a pooling layer, a fully connected layer and an output layer, the feature extraction network includes a feature extraction backbone network and a branch network, the pooling layer includes four sub-pooling layers, and the branch network includes a first branch network, a second branch network and a third branch network; The construction of the initial fish identification model comprises the following steps: Constructing a four-level feature extraction unit to form the feature extraction backbone network, and simultaneously constructing the first branch network, the second branch network, and the third branch network; Sequentially taking the outputs of the feature extraction backbone network, the first branch network, the second branch network, and the third branch network as inputs of the four sub-pooling layers; The outputs of the four sub-pooling layers are used as the inputs of the feature splicing layer, the output of the feature splicing layer is used as the input of the fully connected layer, and the output of the fully connected layer is used as the output layer to obtain an initial fish recognition model.

7. The intelligent fish identification method for fishing in mountain ponds and small reservoirs according to claim 6 is characterized in that: The feature extraction unit satisfies the following relationship: ， Wherein, F is the feature extracted by the feature extraction unit; P is the clear image of the fish school; Indicates that the convolution kernel is 1×1; and The first sub-feature and the second sub-feature are obtained after performing channel segmentation based on the input features of the feature extraction unit; Indicates that the convolution kernel is 3×3 and the expansion rate is Depthwise separable convolution, b=1,2,3,4; Indicates channel shuffling.

8. The intelligent fish identification method for fishing in mountain ponds and small reservoirs according to claim 6 is characterized in that: The branch network satisfies the following relationship: ， in, is the feature extracted by the t-th branch network; Indicates that the convolution kernel is 1×1; is the fusion function, The features extracted by the feature extraction unit at level c, c = 1, 2, 3; Indicates maximum pooling; Indicates that the convolution kernel is 3×3 and the expansion rate is Depthwise separable convolution, d=1,2,3; Indicates upsampling; t=c.

9. The intelligent fish identification method for mountain pond and small reservoir fishing according to claim 5 is characterized in that: The step of grading the fish in the clear image of the fish school according to the recognition result and estimating the proportion of fish of different grades in the fish school comprises the following steps: Based on the identification results, the fish are divided into different grades according to the growth stages of the fish; The average number of fish of different grades in the plurality of clear images of the fish school is calculated, and then the proportion of fish of different grades in the fish school is estimated.

10. An intelligent fish identification system for mountain ponds and small reservoirs, comprising: A data acquisition device, a data output device, a processor and a storage device, wherein the storage device includes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by the processor, the processor implements a fish intelligent identification method for mountain pond and small reservoir fishing as described in any one of claims 1 to 9.

Images