CN-122020415-A - Multi-dimensional characteristic index fusion type intelligent turtle grading method and system

CN122020415ACN 122020415 ACN122020415 ACN 122020415ACN-122020415-A

Abstract

The invention discloses an intelligent turtle grading method and system with multi-dimensional characteristic index fusion, and relates to the technical field of turtle quality grading. Aiming at the problems of strong subjectivity, low efficiency, non-uniform indexes, lack of automation and traceability and the like of the traditional soft-shelled turtle classification, a multi-mode data acquisition and intelligent evaluation system is constructed. The method comprises the steps of synchronously collecting the length of a dorsal scale, the width of a skirt edge, the thickness of a body, the weight and the average value of the body surface chromaticity of a soft-shelled turtle through image measurement, toF detection and precise weighing, calculating composite characteristic parameters after standardization processing, generating a comprehensive quality index (QPI) through a weighted nonlinear model, mapping into standardized scores through a Sigmoid function, and grading. The system has self-learning capability and interpretability output, synchronously generates the traceability two-dimensional code containing acquisition time, place and turtle age, realizes full-automatic, high-precision and nondestructive grading, is suitable for scenes such as farms, sorting centers and the like, and effectively improves the standardization and intellectualization level of turtle grading.

Inventors

WANG YANG
HE HUI
LI SHIYAN
KE QINGQING
ZHOU QIN
LU WEI
Kong Qiaodan
XU ZHENBO
CHEN ZE
Song Shuangying
Ge Yanning
JIA YONGYI
QI MING
LI SIMIN
WANG DINGNAN

Assignees

浙江省水产技术推广总站(浙江省渔业检验检测与疫病防控中心)

Dates

Publication Date: 20260512
Application Date: 20251226

Claims (10)

1. The intelligent turtle grading method with multi-dimensional characteristic index fusion is characterized by comprising the following steps: S1, multi-mode data acquisition, namely synchronously acquiring the length L of the dorsal horn, the width S of the skirt, the body thickness H, the weight M and the body surface chromaticity mean value C of the soft-shelled turtles, wherein the acquisition modes are respectively image automatic measurement, side view or TOF measurement, weighing measurement and RGB image brightness calculation; s2, feature extraction and parameter calculation, namely after standardized processing is carried out on the acquired data, calculating the following composite parameters: Skirt ratio: representing development and growth degree of the skirt, and representing nutritive value; High ratio: Representing the ratio of the height to the length; The fertilizer fullness degree: measuring the reasonable degree of the body weight relative to the body type; Chromaticity correction term: ; Is the standard chromaticity mean; S3, calculating a comprehensive quality index, namely fusing the composite parameters into the comprehensive quality index through a weighted nonlinear comprehensive model, wherein the model expression is as follows: wherein QPI is the integrated quality index, rs is the skirt ratio, rh is the high ratio, fm is the fullness index, ΔC is the chromaticity correction term, rs×Fm is the coupling term, As the weight coefficient of the light-emitting diode, ; S4, grade judgment and output, wherein the comprehensive quality index is mapped into a standardized score of a 0-1 interval through a Sigmoid function, and the mapping formula is as follows: wherein Q is a standardized score, k is an amplification parameter, and θ is a balance parameter; When the grade is I grade A, the grade is output as a high-quality product; when the grade is II grade B, outputting the grade as a standard product; when the grade is III grade C, outputting the grade as a general product; and S5, model self-adaptive optimization, namely automatically fine-tuning a weight coefficient by the model when the manual rechecking sample deviates from the prediction grade, and dynamically switching a corresponding parameter configuration file by the system according to the variety, the cultivation season or the specification of the soft-shelled turtles, wherein an updating strategy adopts incremental fine tuning.
2. The method for grading the intelligent soft-shelled turtle fused by the multidimensional characteristic indexes according to claim 1, wherein in S1, in the process of multi-mode data acquisition, the calibration of the time consistency and the space correspondence of visual images, depth data, weight data and RGB (red, green and blue) chromaticity data is realized through a unified clock signal and a calibration plate.
3. The method for grading soft-shelled turtles with multi-dimensional characteristic index fusion as set forth in claim 1, wherein in S2, said standard chromaticity mean value The method comprises the steps of measuring 30-50 soft-shelled turtle samples which are manually evaluated to be high-quality, calculating the body surface chromaticity mean value of the samples, and averaging.
4. The method for grading soft-shelled turtles with multi-dimensional characteristic index fusion as claimed in claim 1, wherein in S2, the body surface chromaticity mean value extracts target area pixels from RGB images according to a formula And calculating a gray brightness average value, and normalizing the gray brightness average value to a 0-1 interval to obtain the gray brightness average value.
5. The method for grading the intelligent soft-shelled turtle fused with multidimensional feature indexes according to claim 1, wherein in the process of feature extraction and parameter calculation of S2, normalization processing adopts a normalization mode, so that raw data in different dimensions are converted into the same numerical scale.
6. The method for grading soft-shelled turtles with multi-dimensional characteristic index fusion as set forth in claim 1, wherein in S3, weight coefficients are obtained And (3) jointly calibrating the experience of the turtle culture expert and the data of the multiple batches of turtle samples, and adjusting the value of each weight coefficient by taking the minimum grading error as a target in the calibration process.
7. The method for grading the intelligent soft-shelled turtles by fusing multidimensional characteristic indexes according to claim 1, wherein in S4, the amplification parameter k and the balance parameter theta are determined through joint calibration, and parameters are adjusted according to service requirements of soft-shelled turtle quality grading in the calibration process, so that a standardized score has a clear soft-shelled turtle quality evaluation meaning.
8. The method for grading the intelligent soft-shelled turtles by fusing multidimensional characteristic indexes according to claim 1, wherein in the grade judging and outputting process of S4, characteristic values, contribution proportion thereof, model confidence and two-dimensional code tracing information are also output, and the two-dimensional code tracing information comprises acquisition time, location and age of soft-shelled turtles.
9. The method for grading soft-shelled turtles based on multi-dimensional characteristic index fusion according to claim 1, wherein in the model self-adaptive optimization process of S5, when the new data volume is insufficient or the updating effect does not reach a preset standard, the system keeps original model parameters and grading stability is maintained.
10. An intelligent turtle grading system with multi-dimensional characteristic index fusion, which is used for realizing the intelligent turtle grading method with multi-dimensional characteristic index fusion according to any one of claims 1-9, and is characterized by comprising a data acquisition layer, a characteristic calculation layer, a comprehensive evaluation layer and a grading output layer; The data acquisition layer comprises a vision acquisition unit, a ToF depth sensor, a precision weighing module and an annular LED light supplementing and image calibrating assembly, and is used for synchronously acquiring the length of the dorsal scale, the width of the skirt, the body thickness, the weight and the body surface chromaticity average value of the turtles; The characteristic calculation layer is used for carrying out standardized processing on the acquired data, calculating skirt ratio, high ratio, fullness index, chromaticity correction item and coupling item, and outputting standardized characteristic vector; The comprehensive evaluation layer comprises a model training module and a comprehensive quality index calculation module, wherein the model training module determines a weight coefficient, an amplification parameter and a balance parameter through a soft-shelled turtle sample training set, and the comprehensive quality index calculation module calculates a comprehensive quality index based on a feature vector and the weight coefficient and maps the comprehensive quality index to a standardized score; And the grading output layer is used for outputting the turtle grade label, the characteristic contribution proportion, the model confidence and the two-dimensional code traceability information and supporting the docking with a management system or a supervision platform of a turtle breeding enterprise.

Description

Multi-dimensional characteristic index fusion type intelligent turtle grading method and system Technical Field The invention relates to the technical field of intelligent classification of aquatic products, in particular to an intelligent turtle grading method and system with multi-dimensional characteristic index fusion. Background The quality classification of soft-shelled turtles as high-value aquaculture varieties directly influences market pricing, consumer trust and industrial standardized development. Currently, the grading of turtles is mainly performed by artificial sensory evaluation, an operator observes the shape, the width and the body surface color of the waistcoat by experience, and combines the tactile sensation perception state and the weight estimation quality. Although the laboratory physicochemical detection has objectivity, the process is complex, the cost is high, the time consumption is long, destructive sampling is needed, the method cannot be applied to the field real-time detection of living soft-shelled turtles, and the method is difficult to be converted into practical grading application. In recent years, artificial intelligence and machine vision have been applied in the field of agricultural product classification, but soft body surfaces, complex forms and movable living bodies of soft-shelled turtles make quality core indexes difficult to precisely quantify, and in detection, posture change and body surface texture difference easily cause data distortion. At present, a mature commercial intelligent grading scheme for soft-shelled turtles does not appear, and the industry lacks a unified and quantifiable evaluation system and a traceability mechanism, so that the standardization and intelligent upgrading of soft-shelled turtle cultivation industry are restricted. In summary, the existing turtle classification method has the problems of strong subjectivity, different standards, low efficiency, strong destructiveness, lack of automation and traceability and the like, and an intelligent classification technology which can adapt to the characteristics of the turtle species, realize full automation, high precision, no damage and traceability is needed so as to solve the defects of the traditional method and promote the standardized and intelligent development of the turtle cultivation industry. Disclosure of Invention The invention aims to provide an intelligent turtle grading method and system with multi-dimensional characteristic index fusion, which aim to solve the problems of strong subjectivity, low efficiency, non-uniform index, lack of automation and traceability and the like in grading the quality of high-value fresh water products with soft surfaces, complex shapes and movable living bodies of Chinese turtles, soft turtles and the like, and construct a set of full-automatic, high-precision, lossless and interpretable intelligent grading system. In order to achieve the purpose, the invention provides an intelligent turtle grading method with multi-dimensional characteristic index fusion, which is characterized by comprising the following steps: S1, multi-mode data acquisition, namely synchronously acquiring the length L of the dorsal horn, the width S of the skirt, the body thickness H, the weight M and the body surface chromaticity mean value C of the soft-shelled turtles, wherein the acquisition modes are respectively image automatic measurement, side view or TOF measurement, weighing measurement and RGB image brightness calculation; s2, feature extraction and parameter calculation, namely after standardized processing is carried out on the acquired data, calculating the following composite parameters: Skirt ratio: representing development and growth degree of the skirt, and representing nutritive value; High ratio: representing the ratio of body thickness to length; Fullness index: measuring the reasonable degree of the body weight relative to the body type; Chromaticity correction term: ; Is the standard chromaticity mean; S3, calculating a comprehensive quality index, namely fusing the composite parameters into the comprehensive quality index through a weighted nonlinear comprehensive model, wherein the model expression is as follows: wherein QPI is the integrated quality index, rs is the skirt ratio, rh is the high ratio, fm is the fullness index, ΔC is the chromaticity correction term, rs×Fm is the coupling term, As the weight coefficient of the light-emitting diode,; S4, grade judgment and output, wherein the comprehensive quality index is mapped into a standardized score of a 0-1 interval through a Sigmoid function, and the mapping formula is as follows: wherein Q is a standardized score, k is an amplification parameter, and θ is a balance parameter; When the grade is I grade A, the grade is output as a high-quality product; when the grade is II grade B, outputting the grade as a standard product; when the grade is III grade C, outputting the grade as a general