Aquatic seedling counting and growth parameter measuring system

Abstract

The invention discloses an aquatic seedling counting and growth parameter measuring system, which comprises the following steps: acquiring a seedling flow image sequence and generating an image confidence score, selecting kernel function width according to different confidence conditions to construct a density map, and obtaining a density distribution feature vector capable of representing a spatial aggregation state; extracting morphological features of the seedlings under a multi-scale receptive field based on the feature vectors, constructing a multi-scale feature pyramid, calculating a spatial overlapping degree index according to a difference relationship between scales, and identifying a dense region; the fusion weight of the multi-scale features is determined in combination with an overlapping degree index, and fusion features are generated and used for executing counting and growth parameter measurement; and constructing a performance deviation index according to the counting accuracy and the measurement precision, and performing joint updating on the kernel function parameters and the fusion weight when the deviation exceeds a threshold value to form an optimized path of which the parameters can be iteratively updated so as to realize stable counting and accurate measurement under complex density change and overlapping conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent monitoring of aquaculture, and particularly relates to a system for counting and measuring growth parameters of aquatic seedlings. BACKGROUND

[0002] The counting and growth parameter measurement of aquatic seedlings is a basic link in the production of aquatic seedling culture, and is directly related to key processes such as feeding management, breeding density regulation, survival rate evaluation and quality grading. The current manual sampling counting method has problems such as low efficiency, large error and poor repeatability, and manual intervention can easily cause stress to the larvae, affecting the survival rate. Therefore, the automatic counting method based on image processing has gradually become an important direction for intelligent management of aquatic seedling culture.

[0003] With the development of deep learning and computer vision technology, some researches have begun to apply convolutional neural networks to detect and segment fry or shrimp larvae. For example, CN114612397A provides a method for sorting fry by using zoom imaging, image preprocessing, threshold segmentation and target detection model, and then classifying and sorting based on feature information. Although this type of scheme has certain effect on static or medium density scenes, it still has the following technical deficiencies:

[0004] 1. Linear process limits system adaptability. The existing methods generally adopt a linear process of image acquisition, preprocessing, feature extraction and analysis decision, and lack parameter-level and state-level linkage between modules. The processing strategy of the front-end module cannot be dynamically adjusted according to the subsequent results, resulting in performance degradation under complex conditions such as changes in light, fluctuations in density, motion blur or turbidity of water.

[0005] 2. Processing modules are independent and lack a collaborative mechanism. Traditional image preprocessing, local feature extraction and target detection modules are run in an independent manner, and only the results are transmitted between modules, lacking a parameter adjustment mechanism based on feedback. When the density distribution of the seedlings changes greatly, the fixed strategy of preprocessing and feature extraction cannot achieve cross-scene stability.

[0006] 3. Lack of density adaptive processing mechanism. The density of actual seedlings changes significantly in the flow channel, with high overlap of individuals in high-density areas and dispersion of individuals in low-density areas. Existing methods usually use fixed kernel functions or fixed scale models to generate density maps, which are not sensitive enough to density dynamics, and are prone to adhesion errors in high-density areas and missed detection in low-density areas.

[0007] 4. Single feature extraction scale, difficult to balance global and details. The body length of seedlings differs significantly at different growth stages, and the sizes of individuals are uneven. In fixed scale networks, the global contour of large individuals is easily truncated, and the detailed texture features of small individuals are easily lost; the blurred boundaries of overlapping individuals also lead to segmentation errors.

[0008] 5. Insufficient ability to separate overlapping regions: In high-density scenarios, seedling boundaries often stick together, making it difficult for semantic segmentation networks to correctly distinguish overlapping regions. Existing methods generally do not introduce a dedicated overlap evaluation mechanism, resulting in low individual separation accuracy.

[0009] 6. The system lacks a closed-loop optimization mechanism and cannot adaptively update. Most existing systems are open-loop structures with fixed module parameters, making it impossible to automatically learn based on actual operating results. When facing different species, densities, light conditions, and water quality conditions, manual recalibration is required, resulting in high deployment costs and insufficient robustness.

[0010] In view of this, the present invention proposes a system for counting aquatic seedlings and measuring growth parameters. Summary of the Invention

[0011] The purpose of this invention is to provide an aquatic seedling counting and growth parameter measurement system to solve the problems of inaccurate counting, low measurement accuracy, and poor system robustness caused by dynamic changes in density, severe individual overlap, and lack of adaptive mechanisms in the processing flow of existing aquatic seedling counting and growth parameter measurement systems.

[0012] In a first aspect, the present invention provides a system for counting and measuring growth parameters of aquatic seedlings, comprising:

[0013] The image acquisition module is used to acquire seedling flow image sequences and generate image confidence scores based on image sharpness, contrast and noise level;

[0014] An adaptive density map generation module, coupled to the image acquisition module, is used to dynamically select the kernel function width based on the image confidence score to generate a density map and extract the density distribution feature vector.

[0015] The multi-scale feature extraction module, coupled with the adaptive density map generation module, is used to dynamically determine the receptive field range at different scales based on the density distribution feature vector, and extract global morphology, local edge and detail texture features at multiple scales to generate a multi-scale feature pyramid.

[0016] The overlap assessment module, coupled with the multi-scale feature extraction module, is used to identify the seedling boundary contact area based on the multi-scale feature pyramid and calculate the spatial overlap index.

[0017] The feature fusion weight calculation module is deeply coupled with the overlap evaluation module and the multi-scale feature extraction module, and is used to adaptively allocate multi-scale feature fusion weights according to the spatial overlap index.

[0018] The counting and measurement module is used to perform seedling counting and growth parameter measurement based on fused features;

[0019] The closed-loop optimization module, coupled with the adaptive density map generation module, the feature fusion weight calculation module, and the counting and measurement module, is used to adjust the kernel function width and fusion weight in reverse according to the performance deviation index, so as to form a closed-loop feedback path for parameter updates.

[0020] As a preferred embodiment of the present invention, the adaptive density map generation module further includes:

[0021] The image is categorized based on its confidence score, classifying it into high-confidence, medium-confidence, or low-confidence categories; and includes a step of locally modulating the kernel function width based on the local density peaks of the density map, the step comprising:

[0022] Calculate the local density peaks in the neighborhood of each pixel in the density map;

[0023] Compare the local density peak with the preset peak range;

[0024] When the local density peak is in the first interval, the kernel function width is shrunk.

[0025] When the local density peak is in the second interval, the kernel function width is expanded.

[0026] Output the locally modulated kernel function width to ensure consistency between the kernel function width and the local density structure.

[0027] As a preferred embodiment of the present invention, the multi-scale feature extraction module includes calculating the receptive field range at different scales based on the density distribution feature vector; the receptive field range generation steps are as follows:

[0028] Calculate the change in density gradient in the eigenvector of the density distribution;

[0029] The receptive field range of the first scale is determined based on the change in density gradient;

[0030] The receptive field range at the second scale is determined based on the local density dispersion.

[0031] The third-scale receptive field range is determined based on the density peak spacing;

[0032] A consistency check is performed on the three-scale receptive field range to ensure that the generated three-scale receptive field range can be used for the construction of the multi-scale feature pyramid.

[0033] As a preferred embodiment of the present invention, the overlap evaluation module includes a step of performing multi-channel boundary structure analysis on the local edge features in the multi-scale feature pyramid, including:

[0034] Extract the boundary extension length of edge features in the horizontal channel;

[0035] Extract the boundary connectivity patterns of edge features in the vertical channel;

[0036] The boundary contact area is constructed based on the intersection nodes of horizontal and vertical channels;

[0037] Calculate the ratio of the average area of ​​the boundary contact area to the average area of ​​the seedling area;

[0038] The ratio is output as a spatial overlap index for use by the feature fusion weight calculation module.

[0039] As a preferred embodiment of the present invention, the overlap evaluation module further includes a step of constructing a three-dimensional boundary voxelized structure based on the multi-scale feature pyramid to analyze local overlap relationships, the step including:

[0040] Local edge features are mapped to a two-dimensional coordinate grid to form initial boundary segments;

[0041] Voxelized units are constructed by stacking boundary fragments in the vertical direction based on density distribution feature vectors;

[0042] Calculate the common projected area between adjacent voxelized units to form voxel overlap segments;

[0043] Perform a ratio-structured operation on the overlapping voxel fragments and the average area of ​​the seedling region;

[0044] The ratio is output as a supplementary spatial overlap index for use by the feature fusion weight calculation module.

[0045] As a preferred embodiment of the present invention, the feature fusion weight calculation module includes the step of constructing a three-stage scale weight sequence based on a spatial overlap index, the step including:

[0046] The spatial overlap index is classified into three intervals: the first interval, the second interval, and the third interval.

[0047] In the first interval, a weight sequence based primarily on third-scale features is constructed.

[0048] In the second interval, a weight sequence based primarily on second-scale features is constructed.

[0049] In the third interval, a weight sequence based primarily on the first-scale features is constructed.

[0050] Output the complete scale weight sequence for subsequent fusion.

[0051] As a preferred embodiment of the present invention, the feature fusion weight calculation module further includes a weight regression sequence construction step based on regional similarity, the step comprising:

[0052] Map the features at each scale in the multi-scale feature pyramid to a unified feature space;

[0053] Calculate the regional similarity between features at different scales and construct a similarity matrix;

[0054] Perform row normalization on the similarity matrix to form the regression input sequence;

[0055] Calculate the regression weights corresponding to features at each scale based on the regression input sequence;

[0056] The output regression weights serve as supplementary fusion weights for the feature fusion weight calculation module.

[0057] As a preferred embodiment of the present invention, the counting and measurement module includes a step of generating a parameter extraction sequence based on the fusion features, the step including:

[0058] Initial mask is generated by segmenting the seedling region based on fusion features;

[0059] Perform contour structure extraction on the initial mask to determine the body length measurement line;

[0060] Construct a body height measurement area based on the body length measurement line;

[0061] Perform area calculations on the region inside the mask to obtain the area of ​​the local region;

[0062] The output contains a parameter sequence of body length, body height, and region area as growth parameters.

[0063] As a preferred embodiment of the present invention, the counting and measurement module further includes a step of performing measurement based on a hierarchical mask topology, the step comprising:

[0064] Perform a topological layering operation on the initial mask to obtain the main layer and the local layer;

[0065] The principal axis direction is extracted from the main body layer and used as the baseline for body length measurement.

[0066] Extract the local structural contour forming volume height measurement region in the local layer;

[0067] The morphological parameter combination is constructed based on the topological relationship between the main layer and the local layer;

[0068] The output contains a sequence of combined parameters of body length and body height as growth parameter structure terms.

[0069] As a preferred embodiment of the present invention, the closed-loop optimization module further includes the step of constructing a multi-mode joint update path for error samples, the step comprising:

[0070] Density error segments are obtained by performing density map pattern analysis on the error samples.

[0071] Overlap pattern analysis is performed on the error samples to obtain overlapping error segments;

[0072] Construct a kernel function width update sequence based on density error segments;

[0073] Construct a fused weight update sequence based on overlapping error segments;

[0074] The updated sequence is output to simultaneously update the parameters of the adaptive density map generation module and the feature fusion weight calculation module.

[0075] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0076] This invention constructs a deeply coupled collaborative system integrating adaptive density map generation, multi-scale feature extraction, overlap assessment, feature fusion weight calculation, and closed-loop optimization. This system enables the output of upstream modules to directly control the parameters and processing strategies of downstream modules, and allows downstream results to correct upstream parameters in reverse. This forms a dynamic processing mechanism that combines density-driven, overlap-aware, feature-coordinated, and error-feedback mechanisms. Under this mechanism, adaptive density map generation improves the stability of density estimation under varying illumination and noise environments. Multi-scale features, driven by overlap indices, achieve targeted fusion, enhancing the separation capability of overlapping targets. Closed-loop feedback optimization further enables the parameters of each module to automatically converge to the optimal level based on the running results. This significantly improves the system's counting accuracy, overlap separation capability, growth parameter measurement accuracy, and robustness across species and environments in high-density, high-overlap, and multi-dynamic interference scenarios, achieving an overall performance improvement that cannot be achieved by existing linear processes and independent module architectures. Attached Figure Description

[0077] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0078] Figure 1 This is a schematic diagram of the overall architecture of the aquatic seedling counting and growth parameter measurement system of the present invention;

[0079] Figure 2 This is a schematic diagram of the workflow of the adaptive density map generation module of the present invention;

[0080] Figure 3 This is a schematic diagram of the collaborative mechanism for multi-scale feature extraction and fusion in this invention;

[0081] Figure 4 This is a schematic diagram illustrating the coupling relationship between overlap assessment and feature fusion weight calculation in this invention;

[0082] Figure 5 This is a schematic diagram of the feedback adjustment process of the closed-loop optimization module of the present invention. Detailed Implementation

[0083] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below with reference to the accompanying drawings.

[0084] Throughout the accompanying drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions. The described embodiments are only a part of the embodiments of this application, not all of them. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application. The embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0085] Example 1

[0086] Please see Figure 1 As shown, this embodiment provides an aquatic seedling counting and growth parameter measurement system, including an image acquisition module 1, an adaptive density map generation module 2, a multi-scale feature extraction module 3, an overlap evaluation module 4, a feature fusion weight calculation module 5, a counting and measurement module 6, and a closed-loop optimization module 7.

[0087] Image acquisition module 1 is used to acquire seedling flow image sequences and generate image confidence scores based on image quality assessment indicators. Specifically, image acquisition module 1 uses a high-speed CMOS camera with a resolution of 2048×2048 pixels and an adjustable frame rate of 60fps to 120fps. It is equipped with an LED ring light source to provide uniform illumination, with a light source color temperature of 5500K to 6500K and an illuminance of 1000lux to 2000lux. The seedling flow channel width is 200mm to 300mm, and the water depth is controlled at 20mm to 50mm to ensure that the seedlings flow through the imaging area in a single-layer distribution.

[0088] Image quality assessment employs a multi-index fusion strategy. First, the image sharpness index is calculated, and the Tenengrad gradient function is used to evaluate image sharpness.

[0089] ,

[0090] in: For clarity indicators, and These represent the number of rows and columns of the image, respectively. and Each pixel The horizontal and vertical gradients at a given point are calculated using the Sobel operator.

[0091] Next, the image contrast ratio is calculated using the Michelson contrast formula:

[0092] ,

[0093] in: For contrast ratio, and These are the maximum and minimum gray values ​​of the image, respectively.

[0094] The image noise level index was recalculated using the local standard deviation statistical method:

[0095] ,

[0096] in: As an indicator of noise level, This is the number of image blocks (usually between 64 and 256). For the first The standard deviation of pixel grayscale values ​​within an image block.

[0097] Finally, the image confidence score is calculated based on the above three indicators:

[0098] ,

[0099] in: Score the confidence level of the image. , , These are weighting coefficients, taken as 0.5, 0.3, and 0.2 respectively. , , These are reference values, obtained based on statistics from standard samples. The image confidence score ranges from 0 to 1, with higher scores indicating better image quality.

[0100] Images are categorized into three levels based on their confidence scores: high-confidence images (…). Medium confidence images ( ) and low-confidence images ( ).

[0101] The adaptive density map generation module 2 is coupled to the image acquisition module 1. It dynamically adjusts the density map generation strategy based on the image confidence score, generating density maps using different kernel widths for images with different confidence levels, and outputs a density distribution feature vector. (See also...) Figure 2The core function of the adaptive density map generation module 2 is to convert the input image into a density map, where the value of each pixel in the density map represents the seedling density at that location.

[0102] Adaptive density map generation uses a Gaussian kernel function to calculate density values. For the seedling center points detected in the image... It affects surrounding pixels. The density contribution is:

[0103] ,

[0104] in: For pixels Density value at that location, The standard deviation of the Gaussian kernel function (i.e., the kernel width parameter) controls the diffusion range of the density distribution.

[0105] The key innovation of this invention lies in adaptively adjusting the kernel function width based on the image confidence score. For high-confidence images ( (Using a smaller kernel width) to Pixels, preserving fine density distribution information; for medium confidence images ( ), using a medium kernel width to Pixel count, balancing accuracy and robustness; for low-confidence images ( ), using a larger kernel function width to Pixels enhance robustness against noise and blur.

[0106] The adaptive mapping relationship between kernel function width and image confidence score is as follows:

[0107] ,

[0108] The complete density map of the image is obtained by overlaying the density contributions of all seedling center points:

[0109] ,

[0110] in: For pixels The total density value at that location, This represents the number of individual seedlings in the image. For the first The density contribution of each seedling to that pixel.

[0111] After generating the density map, the adaptive density map generation module 2 further extracts the density distribution feature vector. The density distribution feature vector includes five key features: density peak location, density distribution standard deviation, density distribution skewness, density distribution kurtosis, and the location of the maximum density gradient.

[0112] Density peak location Defined as the coordinates of the pixel with the highest density value in the density map:

[0113] ,

[0114] Density distribution standard deviation Reflects the degree of dispersion of the density distribution:

[0115] ,

[0116] in: This represents the average density value of the density map. .

[0117] density distribution skewness Reflecting the asymmetry of density distribution:

[0118] ,

[0119] Density distribution kurtosis Reflecting the sharpness of the density distribution:

[0120] ,

[0121] Location of maximum density gradient Defined as the coordinates of the pixel with the largest gradient magnitude in the density map, this location typically corresponds to the boundary region of an individual seedling:

[0122] ,

[0123] in: and Density map at pixel points The horizontal and vertical gradients at the location.

[0124] The above five features are combined into a density distribution feature vector:

[0125] ,

[0126] The density distribution characteristic vector As the core output of the adaptive density map generation module 2, it is passed to the downstream multi-scale feature extraction module 3 to establish a parameter-level coupling relationship between the two modules.

[0127] The multi-scale feature extraction module 3 is coupled with the adaptive density map generation module 2 to determine the receptive field range for feature extraction based on the density distribution feature vector, extracting seedling morphological features at at least three different scales, and generating a multi-scale feature pyramid. (See also...) Figure 3 The multi-scale feature extraction module 3 adopts an encoder-decoder architecture. The encoder extracts features at three different scales, and the decoder performs feature upsampling and fusion.

[0128] The key innovation of this invention lies in dynamically determining the receptive field range at each scale based on the density distribution feature vector. (Density distribution standard deviation) This parameter reflects the dispersion of seedling distribution and is directly used to adjust the receptive field range.

[0129] First-scale (global scale) receptive field range :

[0130] ,

[0131] in: The base receptive field radius is set to 64 to 128 pixels. This is an adjustment coefficient, ranging from 0.3 to 0.5. When the density distribution dispersion is large, As the size increases, the global receptive field expands accordingly, allowing for the capture of seedling distribution information over a wider spatial range.

[0132] Second-scale (local scale) receptive field range :

[0133] ,

[0134] in: The adjustment factor is 0.2 to 0.3. The receptive field at the local scale is approximately 50% to 70% of the global scale.

[0135] The receptive field range at the third scale (detail scale) :

[0136] ,

[0137] in: The adjustment factor is set to 0.1 to 0.2. The receptive field at the detail scale is approximately 25% to 40% of the global scale.

[0138] After determining the receptive field range at each scale, the multi-scale feature extraction module 3 extracts seedling morphological features at the three scales respectively. A depthwise separable convolutional network is used as the backbone network for feature extraction, with a network depth of 16 to 24 layers, including 5 downsampling stages. After each stage, the feature map resolution is halved and the number of channels is doubled.

[0139] First-scale feature extraction: Feature maps are extracted after the third downsampling stage. The feature map resolution is the same as that of the original image. The number of channels ranges from 256 to 512. This scale feature primarily captures global morphological information of seedlings, including overall outline, orientation, and population distribution patterns.

[0140] Second-scale feature extraction: Extracting feature maps after the second downsampling stage. The feature map resolution is the same as that of the original image. The number of channels ranges from 128 to 256. This scale feature primarily captures local edge information of the seedling, including the head boundary, tail boundary, and body surface contour.

[0141] Third-scale feature extraction: Extracting feature maps after the first downsampling stage. The feature map resolution is the same as that of the original image. The number of channels ranges from 64 to 128. This scale feature primarily captures detailed texture information of the seedlings, including body surface texture, fin details, and eyepoint features.

[0142] Feature maps at three scales , , Constructing a multi-scale feature pyramid:

[0143] ,

[0144] The multi-scale feature pyramid, as the output of the multi-scale feature extraction module 3, is passed to the overlap evaluation module 4 and the feature fusion weight calculation module 5 to establish the coupling relationship between the modules.

[0145] The overlap assessment module 4 is coupled with the multi-scale feature extraction module 3 to calculate the spatial overlap index between individual seedlings based on a multi-scale feature pyramid, identifying the distribution ratio of dense and sparse regions. See also Figure 4 The core function of the overlap assessment module 4 is to analyze the spatial relationship between individual seedlings and quantify the degree of overlap between individuals.

[0146] Overlap assessment is primarily based on second-scale feature maps. This scale is chosen because it strikes a good balance between spatial resolution and semantic information, making it most suitable for boundary analysis. First, the feature map... Channel attention weighting is applied to enhance edge response:

[0147] ,

[0148] in: This is the feature map after edge enhancement. This represents element-wise multiplication. The channel attention weights are calculated using global average pooling and 1×1 convolution.

[0149] Then, the edge enhancement feature map is binarized to obtain the foreground mask of the individual seedling:

[0150] ,

[0151] in: Foreground mask, For indicator functions, The threshold value is determined using the Otsu adaptive thresholding method for binarization.

[0152] Connectivity analysis is performed on the foreground mask to extract each connected region as a candidate seedling individual:

[0153] ,

[0154] in: for A connected region, Indicates the first A set of pixels in a connected region.

[0155] For any two adjacent connected regions and Calculate their spatial overlap:

[0156] ;

[0157] in: For the region and Spatial overlap The area of ​​the intersection of the two regions. and These represent the areas of the two regions. The overlap value ranges from 0 to 1, where 0 indicates that the two regions do not overlap at all, and 1 indicates that the two regions completely overlap.

[0158] Calculate the average spatial overlap index of the entire map:

[0159] ;

[0160] in: This is an index of average spatial overlap. The set of all adjacent region pairs. This represents the number of adjacent region pairs.

[0161] Based on spatial overlap index, image regions are divided into three categories: high overlap regions ( ), overlapping regions ( ) and low overlap regions ( ). Statistics on the area proportion of various regions:

[0162] ,

[0163] in: , , These represent the area percentages of highly overlapping, moderately overlapping, and lowly overlapping regions, respectively. , , The total area of ​​each type of region, This represents the total area of ​​the image.

[0164] The spatial overlap index and regional distribution ratio are output by the overlap assessment module 4 and passed to the feature fusion weight calculation module 5 to establish a deep coupling relationship between the two modules.

[0165] The feature fusion weight calculation module 5 is deeply coupled with the overlap evaluation module 4 and the multi-scale feature extraction module 3, and is used to determine the fusion weights of features at each scale based on the spatial overlap index. The core innovation of this invention lies in the design of an adaptive weight allocation strategy based on overlap, realizing an intelligent mechanism that dynamically adjusts the feature fusion weights according to the overlap.

[0166] For the three scale features in the multi-scale feature pyramid , , Its fusion weight , , Dynamically calculated based on spatial overlap index:

[0167] ,

[0168] in: , , These are the fusion weights for features at the first scale (global scale), the second scale (local scale), and the third scale (detail scale), respectively. This is an index of average spatial overlap. The overlap threshold is set between 0.35 and 0.45. The slope coefficient controls the degree of drastic change in the weights, and is set to 8 to 12.

[0169] The design concept of the above weight calculation formula is as follows:

[0170] when When images are in a state of high overlap, it is necessary to enhance the weight of detailed features to improve individual image separation capabilities. Significantly greater than 1 Approaching 1, When the value approaches 0, the system automatically increases the weight of fine-grained features (third scale) and decreases the weight of coarse-grained features (first scale).

[0171] when When the images are in a low-overlap state, it is necessary to enhance the weights of global features to improve overall recognition stability. Significantly less than 1, Approaching 1, When the value approaches 0, the system automatically increases the weight of coarse-grained features (first scale) and decreases the weight of fine-grained features (third scale).

[0172] when At this point, when the images are in a state of moderate overlap, the weights of the three scale features are nearly evenly distributed. to , to , to .

[0173] Based on the calculated fusion weights, a weighted fusion is performed on the multi-scale feature pyramid. First, the feature maps from the three scales are upsampled to a uniform resolution (usually a fraction of the original image resolution). Then, a weighted summation is performed:

[0174] ,

[0175] in: The fused feature map Indicates an upsampling operation. This indicates a downsampling operation, implemented using bilinear interpolation.

[0176] Fusion Feature Map It incorporates complementary information from multiple scales, preserving rich detailed texture features in highly overlapping regions and stable global morphological features in low-overlapping regions, achieving a synergistic effect of 1+1>2.

[0177] The counting and measurement module 6 performs seedling counting and growth parameter measurement based on fused features. First, the fused feature map... Semantic segmentation is performed to generate segmentation masks for individual seedlings. A fully convolutional network is used as the segmentation network, and the network output is a binary mask with foreground pixels having a value of 1 and background pixels having a value of 0.

[0178] Connectivity analysis is performed on the segmentation mask to extract individual seedlings:

[0179] ,

[0180] in: for Segmentation mask for each individual seedling Indicates the first A collection of pixels for each individual seedling.

[0181] Seedling count results Equal to the number of segmentation masks:

[0182] ,

[0183] For each individual seedling Growth parameters were extracted. These parameters included body length. , body height Head area Tail area and body surface projection area .

[0184] Body surface projection area Equal to the number of pixels in the seedling mask:

[0185] ,

[0186] Body length of individual seedlings Defined as the length of the longest side of the largest bounding rectangle of the individual mask:

[0187] ,

[0188] in: and The first The width and height of the minimum bounding rectangle for each individual seedling.

[0189] Individual height of seedlings Defined as the length of the shorter side of the largest bounding rectangle of this individual mask:

[0190] ,

[0191] The head and tail regions were segmented using a morphological analysis-based method. First, the skeleton lines of the seedling individual mask were calculated, with the two endpoints corresponding to the center of the head and the center of the tail, respectively. Then, a circular region centered at the head center and with a radius of 25% of the body length was designated as the head region, and a circular region centered at the tail center and with a radius of 20% of the body length was designated as the tail region. The head region area... and tail area These represent the number of pixels within the corresponding area.

[0192] Based on the growth parameters of various individual seedlings, the growth evenness index of the population was calculated. Growth evenness was expressed using the coefficient of variation (CV).

[0193] ,

[0194] in: As an indicator of growth uniformity, The standard deviation of body length, This represents the average body length. A smaller coefficient of variation indicates a more uniform population growth; a larger coefficient of variation indicates greater variation in individual size within the population.

[0195] Simultaneously, the skewness and kurtosis of the body length distribution are calculated to reflect the asymmetry and concentration of population growth distribution:

[0196] ,

[0197] Module 6 of the counting and measurement module outputs the seedling count results. Set of growth parameters for each individual organism and population statistics (mean body length) Standard deviation of body length Coefficient of variation skewness kurtosis ).

[0198] The closed-loop optimization module 7 is coupled with the counting and measurement module 6, the feature fusion weight calculation module 5, and the adaptive density map generation module 2. This coupling is used to calculate performance deviation indicators based on counting accuracy and measurement precision, and to adjust the parameters of the preceding modules in reverse according to the characteristics of the error samples, forming a closed-loop feedback path for parameter optimization. (See also...) Figure 5 The closed-loop optimization module 7 enables the system to adapt to learning.

[0199] The core idea of ​​closed-loop optimization is to identify performance bottlenecks based on the accuracy of counting and measurement results, adjust the key parameters of the front-end modules in a targeted manner, and bring the system to the optimal working state through multiple rounds of closed-loop iteration.

[0200] First, the closed-loop optimization module 7 evaluates the counting accuracy. The system counting results are then analyzed. Compared with the actual count marked by humans Compare the results and calculate the counting error rate:

[0201] ,

[0202] in: This is the counting error rate. When... Time (error threshold) Typically, a range of 2% to 5% is taken to indicate that the system's counting accuracy is substandard and parameter optimization is required.

[0203] Secondly, the closed-loop optimization module 7 evaluates the measurement accuracy. This includes measuring the seedling length... Compared with the actual body length measured manually Compare the measurements and calculate the measurement error:

[0204] ,

[0205] in: This represents the average measurement error rate. When... Time (measurement error threshold) Typically, a range of 3% to 6% is taken to indicate that the system's measurement accuracy is substandard and parameter optimization is necessary.

[0206] When the counting error rate or measurement error rate exceeds the threshold, the closed-loop optimization module 7 initiates the error sample analysis process. An error sample is defined as an image or individual whose counting or measurement result deviates significantly from the true value.

[0207] For the counting error samples, analyze their density distribution characteristics. Calculate the difference between the density map of the error samples and the true density map:

[0208] ,

[0209] in: For density map differences, and The predicted density map and the true density map are respectively located at the pixel level. The density value at that location.

[0210] When the density maps show significant differences, it indicates that the kernel function width parameter of the adaptive density map generation module 2 is improperly set and needs adjustment. The closed-loop optimization module 7 adjusts the kernel function width corresponding to the confidence level based on the image confidence score of the error samples.

[0211] ,

[0212] in: This is the adjusted kernel width. The kernel width before adjustment. To adjust the step size, it is usually taken as 1 to 3 pixels. For sign functions. When When the prediction density is too high, it indicates that the kernel function width needs to be reduced; when If the prediction density is too low, it indicates that the kernel function width needs to be increased.

[0213] For measurement error samples, analyze their overlap characteristics. Calculate the spatial overlap index of the region where the error samples are located to identify whether they are in a highly overlapping area. If the error samples are mainly concentrated in highly overlapping areas, it indicates that the feature fusion weight calculation module 5 has insufficient weight for detailed features in highly overlapping scenarios, and the weight coefficient of the third-scale features needs to be increased.

[0214] Closed-loop optimization module 7 adjusts the overlap threshold in the feature fusion weight calculation formula. and slope coefficient :

[0215] ,

[0216] in: This is the threshold adjustment amount, ranging from -0.02 to 0.02. This is the slope adjustment amount, ranging from -1 to 1. The adjustment direction is determined based on the overlap distribution characteristics of the error samples.

[0217] After parameter adjustment, the closed-loop optimization module 7 returns to the system to reprocess the image, calculate new counts and measurement results, and evaluate whether performance has improved. This process forms a complete closed-loop iteration:

[0218] Initial state: The adaptive density map generation module 2 uses the initial kernel function width parameter. The feature fusion weight calculation module 5 uses the initial weight parameters. ;

[0219] First iteration: Image acquisition → Density map generation → Feature extraction → Overlap evaluation → Feature fusion → Counting measurement → Performance evaluation → Identify error samples → Analyze density features and overlap features → Adjust kernel function parameters and weight parameters;

[0220] Second iteration: Reprocess based on the adjusted parameters → Generate new counting measurement results → Evaluate performance improvement → Stop iteration if performance meets the target, otherwise continue adjusting parameters;

[0221] Nth iteration: Repeat the above process until the termination condition is met (performance meets the target or the number of iterations reaches the upper limit, usually 3 to 5 rounds).

[0222] Experimental results show that after 3 to 5 rounds of closed-loop iteration, the system's counting accuracy improved from the initial 85% to 88% to over 98%, and the measurement error decreased from the initial 7% to 9% to below 3%, demonstrating a significant performance improvement.

[0223] Example 2:

[0224] To verify the technical effects of this invention, a counting and measurement experiment based on cyprinid fish fry was conducted at an aquatic seedling farm. The experimental subjects were koi and crucian carp fry, which were in the growth stage of 7 to 15 days after hatching and had a body length ranging from 8 mm to 25 mm.

[0225] Experimental equipment: high-speed CMOS camera (2048×2048 pixels, frame rate 120fps), LED ring light source (illuminance 1500 lux, color temperature 6000K), seedling flow channel (width 250mm, water depth 30mm, flow rate 20mm / s to 50mm / s).

[0226] Experimental Method: 100 sets of images of seedling movement were collected, each set containing 50 to 200 individual seedlings. The number and length of the seedlings were manually and precisely labeled as the true values. The system of this invention was used for automatic counting and measurement. The system results were compared with the manually labeled values ​​to calculate the counting accuracy and measurement error.

[0227] Experimental results:

[0228] Initial state (closed-loop optimization not enabled): counting accuracy 87.3%, measurement error 7.8%.

[0229] After the first round of closed-loop iteration: the counting accuracy was 91.5%, the measurement error was 5.6%, and the performance was significantly improved.

[0230] After the second round of closed-loop iteration: the counting accuracy was 95.2%, the measurement error was 4.1%, and continuous optimization was achieved.

[0231] After the third round of closed-loop iteration: the counting accuracy was 98.1% and the measurement error was 2.9%, reaching a high level of precision.

[0232] Comparative experiment: Using the existing technical solution (the method in CN114612397A) to process the same dataset, the counting accuracy was 84.6%, and the measurement error was 9.3%. Compared with the existing technology, the present invention improves the counting accuracy by 13.5 percentage points and reduces the measurement error by 6.4 percentage points, demonstrating significant technical advantages.

[0233] To verify the robustness of this invention under different density scenarios, three sets of density gradient experiments were designed:

[0234] Low-density scenes: Each frame contains 20 to 50 seedlings with a spacing of more than 50 mm and virtually no overlap.

[0235] Medium density scene: Each frame contains 60 to 120 seedlings with a spacing of 20mm to 50mm between seedlings, and some individuals have slight overlap (overlap rate of 10% to 30%).

[0236] High-density scenes: Each frame contains 150 to 250 seedlings with a spacing of less than 20 mm, and most individuals have moderate to severe overlap (overlap rate of 30% to 70%).

[0237] Experimental results:

[0238] In low-density scenarios: counting accuracy is 99.2%, and measurement error is 2.1%. The adaptive density map generation module uses a narrow kernel function (σ=10 pixels), and the feature fusion weights are biased towards global features (w1=0.5, w2=0.3, w3=0.2).

[0239] Medium density scene: counting accuracy 97.8%, measurement error 3.2%. The adaptive density map generation module uses a medium kernel function (σ=18 pixels) and the feature fusion weights are evenly distributed (w1=0.35, w2=0.35, w3=0.3).

[0240] High-density scenes: counting accuracy of 96.4%, measurement error of 4.5%. The adaptive density map generation module adopts a wide kernel function (σ=28 pixels), and the feature fusion weights are biased towards detailed features (w1=0.2, w2=0.3, w3=0.5).

[0241] Comparative experiments: Existing technologies achieve an accuracy rate of 88.5% in low-density scenarios, 82.3% in medium-density scenarios, and only 71.8% in high-density scenarios, with performance declining sharply as density increases. This invention maintains an accuracy rate of over 96% in all three density scenarios, demonstrating significantly superior robustness compared to existing technologies.

[0242] Experimental analysis shows that the adaptive density map generation and feature fusion weight dynamic adjustment mechanism of the present invention can automatically optimize the processing strategy according to the actual density distribution and overlap, and maintain stable high-precision performance in low-density, medium-density and high-density scenarios, verifying the effectiveness of the module coupling collaborative system.

[0243] To verify the cross-species adaptability of this invention, tests were conducted on five different aquatic seedlings:

[0244] Cyprinid fish (koi, crucian carp): Body length 10mm to 25mm, slender body, oval in side view.

[0245] Perch (California bass, sea bass): Body length 12mm to 28mm, fusiform body shape, large head.

[0246] Grouper family (blue grouper, tiger grouper): 8mm to 20mm in length, short and stout body, with spots on the body surface.

[0247] Prawns (Litopenaeus vannamei, Macrobrachium rosenbergii): Body length 5mm to 15mm, curved body shape, with a distinct tail fan.

[0248] Shellfish (scallops, abalone larvae): 2mm to 8mm in length, irregular in shape, and highly transparent.

[0249] Experimental results:

[0250] Cyprinid fish: Counting accuracy was 98.1%, measurement error was 2.9%, and optimal performance was achieved after 3 rounds of closed-loop iteration.

[0251] Perch family: Counting accuracy 97.6%, measurement error 3.4%, achieving optimal performance after 4 rounds of closed-loop iteration.

[0252] Grouper family: Counting accuracy 96.8%, measurement error 4.1%, achieving optimal performance after 4 rounds of closed-loop iteration.

[0253] Shrimp: Counting accuracy was 97.2%, measurement error was 3.7%, and optimal performance was achieved after 5 rounds of closed-loop iteration.

[0254] Shellfish: Counting accuracy of 95.3%, measurement error of 5.6%, and the best performance was achieved after 5 rounds of closed-loop iteration (shellfish are the most difficult to count due to their high transparency and irregular shape).

[0255] Comparative experiment: The existing technology achieved an average accuracy of 78.5% on five types of seedlings, with large performance fluctuations (standard deviation 11.2%) among different species, requiring parameter recalibration for each type of seedling. The present invention achieved an average accuracy of 97.0% on five types of seedlings, with small performance fluctuations (standard deviation 1.1%), and automatically adapts to different species through a closed-loop optimization mechanism, eliminating the need for manual parameter recalibration.

[0256] Experimental analysis shows that the closed-loop optimization mechanism of this invention enables the system to have adaptive learning capabilities, automatically identify the characteristic differences of different types of seedlings, and adjust the module parameters to the optimal working state through 3 to 5 rounds of closed-loop iteration, thus achieving high adaptability and high robustness across species.

[0257] In summary, this invention achieves high-precision counting and accurate measurement of aquatic seedlings by constructing a deeply coupled and collaborative system of five core modules: adaptive density map generation, multi-scale feature extraction, overlap evaluation, feature fusion weight calculation, and closed-loop optimization. The parameter-level coupling, state-level linkage, and logic-level collaboration between modules form a complete closed-loop feedback optimization path, enabling the system to have adaptive learning capabilities. Experimental verification shows that this invention significantly improves counting accuracy, measurement precision, robustness, and adaptability compared to existing technologies, providing an efficient, accurate, and intelligent automated measurement tool for modern aquatic seedling production.

[0258] Example 3

[0259] To verify the actual effect of the present invention in high-density and complex scenarios, three different environmental conditions (light fluctuation ±30%, water turbidity increased by 40%, and seedling density increased to 150% of the control group) were selected as test scenarios, and the existing open-loop method with fixed parameters and the deep-coupled closed-loop system of the present invention were compared respectively.

[0260] In scenarios with varying natural light and motion blur, this invention employs an adaptive density map generation strategy using narrow, medium, and wide kernels. Test results show that, compared to the fixed kernel function method, the density estimation average error achieved by this invention is reduced by 25% to 35%, and the matching degree between the density distribution standard deviation and the true distribution is improved by 22%.

[0261] In scenarios with an overlap rate of 30%–70%, the individual boundary separation capability is significantly improved through dynamic adjustment of three-scale weights driven by overlap.

[0262] Moderate overlap (approximately 40%–60%): Separation accuracy improved from 75% for the control method to 92%;

[0263] High overlap (approximately 60%–80%): Separation accuracy improved from 60% to 85%.

[0264] In mixed seedling scenarios with significant size differences, the accuracy rate of detection for both large and small individuals remains above 95%.

[0265] This invention constructs a closed-loop structure of density map generation → feature extraction → overlap evaluation → feature fusion → count measurement → performance deviation calculation → parameter feedback, which achieves stable convergence after 3 to 5 iterations, specifically as follows:

[0266] The counting accuracy improved from the initial 85% to over 98%;

[0267] The measurement error of growth parameters was reduced from 8% to below 3%;

[0268] Performance fluctuations under different light, density, and variety conditions do not exceed 3%.

[0269] Comparative experiments show that:

[0270] The accuracy rate using the adaptive density map generation module alone is 88%.

[0271] The accuracy rate of multi-scale feature extraction alone is 90%;

[0272] After coupling the two, the accuracy rate increased to 96%;

[0273] After incorporating overlap assessment and feature fusion, the overall accuracy reaches over 98%, and the processing efficiency and robustness are improved by over 50%.

[0274] On a continuous flow aquaculture seedling production line, the average processing time for a single frame image by the system of this invention is 50-80ms, which can achieve real-time monitoring of 12-20fps, meeting the requirements for online counting and real-time grading processing.

[0275] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A system for counting and measuring growth parameters of aquatic seedlings, characterized in that, include: Image acquisition module (1) is used to acquire seedling flow image sequences and generate image confidence scores based on image clarity, contrast and noise level; An adaptive density map generation module (2) is coupled to the image acquisition module (1) and is used to dynamically select the kernel function width based on the image confidence score to generate a density map and extract the density distribution feature vector. The multi-scale feature extraction module (3) is coupled with the adaptive density map generation module (2) to dynamically determine the receptive field range at different scales based on the density distribution feature vector, and extract global morphology, local edge and detail texture features at multiple scales to generate a multi-scale feature pyramid. The overlap assessment module (4) is coupled with the multi-scale feature extraction module (3) to identify the seedling boundary contact area based on the multi-scale feature pyramid and calculate the spatial overlap index. The feature fusion weight calculation module (5) is deeply coupled with the overlap evaluation module (4) and the multi-scale feature extraction module (3) to adaptively allocate multi-scale feature fusion weights based on the spatial overlap index. The counting and measurement module (6) is used to perform seedling counting and growth parameter measurement based on fusion features; The closed-loop optimization module (7) is coupled with the adaptive density map generation module (2), the feature fusion weight calculation module (5), and the counting and measurement module (6) to adjust the kernel function width and fusion weight in reverse according to the performance deviation index, so as to form a closed-loop feedback path for parameter update.

2. The aquatic seedling counting and growth parameter measurement system according to claim 1, characterized in that, The adaptive density map generation module (2) further includes: The image is categorized based on its confidence score, classifying it into high-confidence, medium-confidence, or low-confidence categories; and includes a step of locally modulating the kernel function width based on the local density peaks of the density map, the step comprising: Calculate the local density peaks in the neighborhood of each pixel in the density map; Compare the local density peak with the preset peak range; When the local density peak is in the first interval, the kernel function width is shrunk. When the local density peak is in the second interval, the kernel function width is expanded. Output the locally modulated kernel function width to ensure consistency between the kernel function width and the local density structure.

3. The aquatic seedling counting and growth parameter measurement system according to claim 2, characterized in that, The multi-scale feature extraction module (3) includes calculating the receptive field range at different scales based on the density distribution feature vector; the steps for generating the receptive field range are as follows: Calculate the change in density gradient in the eigenvector of the density distribution; The receptive field range of the first scale is determined based on the change in density gradient; The receptive field range at the second scale is determined based on the local density dispersion. The third-scale receptive field range is determined based on the density peak spacing; A consistency check is performed on the three-scale receptive field range to ensure that the generated three-scale receptive field range can be used for the construction of the multi-scale feature pyramid.

4. The aquatic seedling counting and growth parameter measurement system according to claim 3, characterized in that, The overlap evaluation module (4) includes a step of performing multi-channel boundary structure parsing on the local edge features in the multi-scale feature pyramid, including: Extract the boundary extension length of edge features in the horizontal channel; Extract the boundary connectivity patterns of edge features in the vertical channel; The boundary contact area is constructed based on the intersection nodes of horizontal and vertical channels; Calculate the ratio of the average area of ​​the boundary contact area to the average area of ​​the seedling area; The ratio is output as a spatial overlap index for use by the feature fusion weight calculation module.

5. The aquatic seedling counting and growth parameter measurement system according to claim 1, characterized in that, The overlap evaluation module (4) further includes a step of constructing a three-dimensional boundary voxelized structure based on the multi-scale feature pyramid to resolve local overlap relationships, the step including: Local edge features are mapped to a two-dimensional coordinate grid to form initial boundary segments; Voxelized units are constructed by stacking boundary fragments in the vertical direction based on density distribution feature vectors; Calculate the common projected area between adjacent voxelized units to form voxel overlap segments; Perform a ratio-structured operation on the overlapping voxel fragments and the average area of ​​the seedling region; The ratio is output as a supplementary spatial overlap index for use by the feature fusion weight calculation module.

6. The aquatic seedling counting and growth parameter measurement system according to claim 5, characterized in that, The feature fusion weight calculation module (5) includes the step of constructing a three-stage scale weight sequence based on the spatial overlap index, the step including: The spatial overlap index is classified into three intervals: the first interval, the second interval, and the third interval. In the first interval, a weight sequence based primarily on third-scale features is constructed. In the second interval, a weight sequence based primarily on second-scale features is constructed. In the third interval, a weight sequence based primarily on the first-scale features is constructed. Output the complete scale weight sequence for subsequent fusion.

7. The aquatic seedling counting and growth parameter measurement system according to claim 6, characterized in that, The feature fusion weight calculation module (5) further includes a weight regression sequence construction step based on regional similarity, the step including: Map the features at each scale in the multi-scale feature pyramid to a unified feature space; Calculate the regional similarity between features at different scales and construct a similarity matrix; Perform row normalization on the similarity matrix to form the regression input sequence; Calculate the regression weights corresponding to features at each scale based on the regression input sequence; The output regression weights serve as supplementary fusion weights for the feature fusion weight calculation module.

8. The aquatic seedling counting and growth parameter measurement system according to claim 1, characterized in that, The counting and measurement module (6) includes a step of generating a parameter extraction sequence based on the fusion features, the step including: Initial mask is generated by segmenting the seedling region based on fusion features; Perform contour structure extraction on the initial mask to determine the body length measurement line; Construct a body height measurement area based on the body length measurement line; Perform area calculations on the region inside the mask to obtain the area of ​​the local region; The output contains a parameter sequence of body length, body height, and region area as growth parameters.

9. The aquatic seedling counting and growth parameter measurement system according to claim 8, characterized in that, The counting and measurement module (6) further includes a step of performing measurement based on a hierarchical mask topology, the step including: Perform a topological layering operation on the initial mask to obtain the main layer and the local layer; The principal axis direction is extracted from the main body layer and used as the baseline for body length measurement. Extract the local structural contour forming volume height measurement region in the local layer; The morphological parameter combination is constructed based on the topological relationship between the main layer and the local layer; The output contains a sequence of combined parameters of body length and body height as growth parameter structure terms.

10. The aquatic seedling counting and growth parameter measurement system according to claim 9, characterized in that, The closed-loop optimization module (7) includes the step of constructing a multi-mode joint update path for error samples, the step including: Density error segments are obtained by performing density map pattern analysis on the error samples. Overlap pattern analysis is performed on the error samples to obtain overlapping error segments; Construct a kernel function width update sequence based on density error segments; Construct a fused weight update sequence based on overlapping error segments; The output update sequence is used to simultaneously update the parameters of the adaptive density map generation module and the feature fusion weight calculation module.

Images