Citrus internal and external quality detection and grading method and system based on deep learning

Abstract

The invention relates to the technical field of quality detection and grading, in particular to a citrus internal and external quality detection and grading method and system based on deep learning. The method comprises the following steps: carrying out spectral image acquisition on oranges to obtain spectral characteristic wave band data; performing tactile perception data acquisition on the citrus to obtain citrus tactile perception data; according to the spectral characteristic wave band data, sugar-acid ratio gradient identification of different areas of the citrus is carried out, internal maturity grade division of the different areas of the citrus is carried out, and internal maturity grade data of the citrus is obtained; performing external form perception analysis of different maturity areas on the citrus internal maturity grade data to obtain citrus external form perception data; performing citrus internal and external quality memory learning on the citrus external form perception data and the citrus internal maturity grade data to obtain citrus internal and external quality memory normalized data; according to the method, the quality detection and grading technology is optimized, so that the quality detection and grading technology is more perfect.

Description

Technical Field

[0001] The present invention relates to the technical field of quality inspection and grading, and in particular to a method and system for inspecting and grading the internal and external quality of citrus based on deep learning. Background Art

[0002] Deep learning technology builds models such as convolutional neural networks (CNNs) and uses a large number of citrus images and corresponding quality data for training to achieve automatic detection of the internal and external quality of citrus. The method first uses an image acquisition system to obtain the appearance image of citrus, and extracts surface features such as color, texture and defects through image processing technology; at the same time, non-destructive detection technology (such as near-infrared spectroscopy) is used to collect citrus internal quality data, such as sugar-acid ratio, internal structural abnormalities, etc. The deep learning model can construct complex nonlinear relationships through the analysis and learning of these data, and achieve efficient and accurate citrus quality grading. Compared with traditional methods, the detection and grading method based on deep learning has higher accuracy, scalability and automation, can reduce human intervention, improve grading efficiency, and support large-scale application in citrus production lines, promoting the digitalization and intelligent transformation of agriculture. However, a traditional method for detecting and grading the internal and external quality of citrus based on deep learning has the problems of inaccurate recognition of the internal and external maturity of citrus and low automation. Summary of the invention

[0003] Based on this, it is necessary to provide a method and system for detecting and grading the internal and external quality of citrus based on deep learning to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above purpose, a method for detecting and grading the internal and external quality of citrus based on deep learning is provided, the method comprising the following steps:

[0005] Step S1: collecting spectral images of citrus through a hyperspectral imager to obtain spectral characteristic band data; collecting tactile perception data of citrus through a tactile sensor built into a mechanical claw to obtain tactile perception data of citrus;

[0006] Step S2: identifying the sugar-acid ratio gradient of different regions of citrus according to the spectral characteristic band data, and obtaining the sugar-acid ratio gradient data of citrus; evaluating the difference in pulp elasticity of citrus in different regions according to the sugar-acid ratio gradient data of citrus and the tactile perception data of citrus, and obtaining the difference in pulp elasticity of citrus; dividing the internal maturity level of different regions of citrus based on the sugar-acid ratio gradient data of citrus and the citrus pulp elasticity prediction data, and obtaining the internal maturity level data of citrus;

[0007] Step S3: performing external morphological perception analysis of different maturity areas on the internal maturity grade data of the citrus according to the spectral characteristic band data to obtain external morphological perception data of the citrus;

[0008] Step S4: Based on the proximal strategy optimization algorithm, the external morphology perception data of the citrus and the internal maturity grade data of the citrus are memorized and learned to obtain the normalized internal and external quality data of the citrus; based on the normalized internal and external quality memory data of the citrus, the internal and external quality selection firmware of the citrus is designed to obtain the internal and external quality selection firmware of the citrus, and the internal and external quality selection firmware of the citrus is sent to the cloud platform to execute the internal and external quality detection and grading of the citrus.

[0009] The present invention collects spectral images and tactile perception data of citrus through a hyperspectral imager and a tactile sensor, greatly improving the accuracy and comprehensiveness of quality detection. The hyperspectral imager can obtain detailed spectral features of the surface of citrus, and these spectral data can reveal the chemical composition and state of the fruit, especially the content of sugar and acidity. This process can effectively identify the characteristics of the fruit at different stages of maturity through the analysis of different bands. In addition, the built-in tactile sensor obtains the hardness and elasticity data of citrus through contact, supplementing the physical properties that the spectral imaging cannot fully reflect. This combination of multidimensional data provides a solid foundation for subsequent classification, so that the quality detection of citrus is not limited to visual features, but comprehensively considers the chemical composition and physical properties, so that the maturity and flavor characteristics of the fruit can be more accurately evaluated. The sugar-acid ratio gradient of different regions of citrus is identified by using spectral feature band data, which is an important basis for quality classification. The sugar-acid ratio is an important indicator for evaluating the taste of fruit, which directly affects consumer choice and market acceptance. By performing gradient identification on the ratios of different regions, it is possible to clarify which regions have sweeter or more sour fruits, thereby classifying them into different consumer markets or usage purposes. In addition, the tactile perception data of citrus is combined with the evaluation of the difference in pulp elasticity, which provides a more comprehensive perspective on the internal quality of the fruit. Fruits with better elasticity usually mean higher water content and better taste, which is crucial for marketing. The comprehensive evaluation at this stage makes the classification of citrus more scientific and provides data support for subsequent sales strategies. By analyzing the external morphological perception of the internal maturity grade data of citrus, a more comprehensive understanding of the relationship between the apparent characteristics and the intrinsic quality of the fruit can be obtained. Using spectral feature band data, this process not only focuses on the physical characteristics of the fruit such as color, size and shape, but also infers the internal maturity through these external morphological characteristics. This method allows researchers and fruit farmers to conduct a systematic analysis of citrus in different maturity areas, so as to formulate corresponding picking and sales strategies. Through external morphological perception data, the health status, maturity and suitable market positioning of the fruit can be quickly identified, which improves the efficiency and accuracy of classification. In addition, this step provides a scientific basis for subsequent quality evaluation and selection, ensuring that consumers can obtain better quality citrus products, thereby improving the efficiency and benefits of the entire supply chain. The process of memorizing and learning the internal and external quality data of citrus based on the proximal strategy optimization algorithm marks a major advancement in fruit quality detection and grading. By combining external morphological perception data with internal maturity grade data, the algorithm can identify the key factors affecting citrus quality and form an efficient quality memory model. This model can normalize different data so that each feature has a consistent standard for subsequent analysis. Based on these normalized data, the designed citrus internal and external quality selection firmware can intelligently identify and classify fruits of different qualities, further improving the automation and intelligence level of detection.This method not only improves the detection efficiency, but also reduces the interference of human factors, ensuring that each citrus can be evaluated scientifically and objectively. By combining the selection firmware with the cloud platform, data can be uploaded and analyzed in real time, providing a powerful data support and feedback mechanism for the entire industry, and ultimately promoting the intelligent and modern process of citrus quality management. Therefore, the present invention provides an optimization process for a traditional method for detecting and grading the internal and external quality of citrus based on deep learning, solves the problems of inaccurate recognition of the internal and external maturity of citrus and low degree of automation in a traditional method for detecting and grading the internal and external quality of citrus based on deep learning, improves the accuracy of identifying the internal and external maturity of citrus, and improves the level of automation.

[0010] Preferably, step S1 comprises the following steps:

[0011] Step S11: collecting spectral images of citrus using a hyperspectral imager to obtain spectral images of citrus;

[0012] Step S12: performing image enhancement processing on the citrus spectrum image to obtain a citrus spectrum enhanced image;

[0013] Step S13: extracting characteristic bands from the citrus spectrum enhanced image to obtain spectral characteristic band data;

[0014] Step S14: collecting tactile perception data of the citrus through the tactile sensor built into the mechanical claw to obtain tactile perception data of the citrus.

[0015] The present invention uses a hyperspectral imager to collect spectral images of citrus and generates detailed spectral images of citrus. This process can capture the reflection characteristics of citrus in different bands, providing rich spectral information for subsequent analysis. The application of hyperspectral imaging technology enables researchers to obtain more spectral data than traditional imaging technology. These data not only reveal the chemical composition, maturity and health of citrus, but also help identify subtle differences related to fruit quality. This preliminary step lays a solid foundation for subsequent quality inspection, classification and maturity assessment, ensuring the scientificity and reliability of subsequent analysis. Image enhancement processing is performed on citrus spectral images to improve the contrast and clarity of the image and make the spectral features more obvious. By using various image processing techniques, such as histogram equalization, filtering and sharpening, noise and blur can be effectively removed, highlighting the subtle spectral changes on the surface of citrus. This processing not only improves the quality of the image, but also enhances the accuracy of subsequent feature extraction. The enhanced spectral image can more clearly show the surface characteristics of the fruit, providing more reliable data support for scientific analysis. In addition, image enhancement also helps researchers to more intuitively identify fruits of different maturity and quality grades, thereby achieving higher accuracy and efficiency in quality assessment. Feature band extraction of citrus spectral enhanced images is a key step in the data analysis process. By analyzing the reflection intensity of different bands in the spectral image, the feature bands related to the internal components of the fruit (such as sugar, acidity, etc.) can be identified. This process effectively simplifies a large amount of spectral information into operational feature data, making subsequent analysis and classification more efficient. Feature band extraction not only improves the efficiency of data processing, but also enhances the understanding of fruit quality, and can accurately distinguish between citrus of different maturity and quality. The results of this step provide a scientific basis for the quality detection and classification of citrus, which helps to improve the management level and market decision-making ability of fruit farmers. The tactile sensor built into the mechanical claw is used to collect tactile perception data of citrus. This process supplements the spectral data and forms a comprehensive quality assessment of the fruit. The tactile sensor can obtain its surface hardness, elasticity and other physical properties by contacting the citrus. These data, combined with spectral information, provide an important supplement to the internal maturity and quality of the fruit. The fusion of this multimodal data makes the quality assessment of citrus more accurate and comprehensive. The results of tactile perception can not only reflect the degree of ripeness of the fruit, but also the degree of

[0016] Preferably, step S2 comprises the following steps:

[0017] Step S21: using a preset spectral band similarity recognition model to classify the spectral feature band data into similar bands to obtain spectral feature band similarity data;

[0018] Step S22: identifying the sugar-acid ratio gradient of different regions of citrus according to the similarity data of the spectral characteristic bands, and obtaining the sugar-acid ratio gradient data of citrus;

[0019] Step S23: evaluating the difference in pulp elasticity of citrus in different areas according to the citrus sugar-acid ratio gradient data and the citrus tactile perception data, and obtaining citrus pulp elasticity difference data;

[0020] Step S24: Based on the citrus sugar-acid ratio gradient data and the citrus pulp elasticity difference data, the internal maturity grade of different citrus areas is divided to obtain the citrus internal maturity grade data.

[0021] The present invention uses a preset spectral band similarity recognition model to classify similar bands of spectral feature band data, aiming to identify bands with similar characteristics through cluster analysis. This process not only improves the efficiency of data processing, but also simplifies the complex spectral data into an operational similar data set. By analyzing the similarity between different bands, researchers can quickly identify key bands related to fruit quality, so as to better understand the relationship between spectral data and the internal components of citrus. The results of similar band classification lay the foundation for the subsequent identification of sugar-acid ratio and analysis of internal maturity, making the evaluation process more accurate. At the same time, this process also provides an important reference for subsequent model optimization and feature selection, ensuring the scientificity and reliability of data analysis. The sugar-acid ratio gradient of different regions of citrus is identified based on the similarity data of spectral feature bands. This identification process can reveal the maturity and taste characteristics of fruits in different regions, and the sugar-acid ratio is an important indicator affecting the flavor of citrus. By analyzing the characteristic bands in the spectral data, researchers can accurately quantify the sugar and acidity of each region, thereby realizing the gradient identification of the sugar-acid ratio. This process not only provides an objective basis for fruit quality assessment, but also evaluates the difference in pulp elasticity in different regions based on the citrus sugar-acid ratio gradient data and citrus tactile perception data. This evaluation process not only supplements the analysis of the intrinsic quality of the fruit, but also reflects the maturity of the fruit through physical properties. The difference in pulp elasticity is directly related to the moisture content, hardness and taste of the fruit. By combining the sugar-acid ratio with tactile perception data, researchers can accurately identify which areas have more elastic flesh and which areas are overripe or immature. This comprehensive evaluation helps ensure the optimal edible state of each fruit, while providing fruit farmers with practical advice on picking and handling, further improving the market value of the fruit and consumer satisfaction. The internal maturity level of different citrus regions is divided based on the citrus sugar-acid ratio gradient data and pulp elasticity difference data, aiming to systematically evaluate the overall quality of the fruit. By comprehensively considering the sugar-acid ratio and pulp elasticity, this division can accurately divide citrus into different maturity levels.

[0022] Preferably, step S22 includes the following steps:

[0023] Step S221: extracting sugar and acid bands of different regions of citrus from similar data of spectral characteristic bands to obtain sugar and acid spectral band data;

[0024] Step S222: performing multi-region sugar-acid distribution chromatography processing on different regions of citrus based on the sugar-acid spectral band data to obtain multi-region sugar-acid distribution chromatography data;

[0025] Step S223: performing band density clustering calculation on the sugar-acid spectrum band data of different regions of citrus according to the multi-region sugar-acid distribution chromatography data to obtain sugar-acid distribution band density clustering data;

[0026] Step S224: identifying the sugar-acid ratio gradients of different citrus regions using the sugar-acid distribution band density clustering data to obtain citrus sugar-acid ratio gradient data.

[0027] The present invention extracts sugar and acid bands of different regions of citrus from similar data of spectral characteristic bands, which is the basic link for analyzing the sugar and acid content of fruits. By selecting specific bands, researchers can effectively extract spectral information related to sugar and acidity, thereby obtaining sugar and acid spectral band data. This extraction process not only improves the accuracy of subsequent analysis, but also ensures that the sugar and acid characteristics of different regions can be reasonably displayed. Accurate sugar and acid band extraction provides important data support for analyzing the flavor characteristics of citrus, allowing researchers to more clearly understand the differences in sugar and acid distribution between different regions. This not only helps to evaluate the maturity and quality of fruits, but also provides a scientific basis for fruit farmers to optimize planting and picking strategies. Multi-region sugar and acid distribution chromatography of different regions of citrus is performed based on sugar and acid spectral band data, aiming to reveal the distribution characteristics of sugar and acid in each region of the fruit. Through chromatography, researchers can visualize sugar and acid data in different regions, thereby forming multi-region sugar and acid distribution chromatography data. This process not only improves the understanding of the distribution of sugar and acid components, but also provides an intuitive basis for subsequent analysis. Through chromatography technology, researchers can identify differences in sugar and acid distribution of fruits under different growth environments or conditions, thus providing an important reference for fruit growth management. This precise regional distribution information can help fruit farmers adjust fertilization and irrigation strategies, thereby improving the overall quality and market value of the fruit. Based on the multi-region sugar and acid distribution chromatography data, the sugar and acid spectral band data of different regions of citrus are clustered and calculated. This process aims to analyze the characteristics of sugar and acid distribution in different regions. Through clustering calculation, researchers can classify similar band data into one category to form sugar and acid distribution band density clustering data. This step not only helps to identify the aggregation trend of sugar and acid in different regions, but also provides an important basis for subsequent flavor evaluation and quality analysis. The results of cluster analysis enable researchers to more intuitively identify which regions have higher sugar and acid content and which regions have relatively uniform sugar and acid distribution. This analysis process enhances the understanding of citrus flavor characteristics and lays a solid foundation for optimizing cultivation technology and improving fruit quality. The sugar and acid ratio gradient of different regions of citrus is identified by clustering data of sugar and acid distribution band density. The core of this process is to achieve quantitative evaluation of the sugar and acid ratio of fruits. By analyzing the cluster data, the researchers were able to accurately identify the sugar-acid ratio gradient in each region, thus forming the sugar-acid ratio gradient data for citrus. This step not only helps to gain a deeper understanding of the taste characteristics of the fruit, but also provides an important basis for the evaluation of the fruit's maturity and flavor.

[0028] Preferably, step S23 includes the following steps:

[0029] Step S231: calculating the elastic modulus of different regions of the citrus based on the tactile perception data of the citrus to obtain regional elastic modulus data of the citrus;

[0030] Step S232: extracting the surface temperature and humidity of different areas of the citrus from the tactile perception data of the citrus to obtain the surface temperature and humidity data of the citrus area;

[0031] Step S233: Calculate the elastic tension of different citrus regions according to the elastic modulus data of the citrus regions and the surface temperature and humidity data of the citrus regions to obtain the elastic tension data of the citrus regions;

[0032] Step S234: performing periodic deposition prediction of different citrus regions on the citrus sugar-acid ratio gradient data to obtain citrus sugar-acid ratio periodic deposition data;

[0033] Step S235: Evaluate the difference in pulp elasticity of different regions of citrus according to the citrus sugar-acid ratio periodic deposition data and the citrus regional elastic tension data to obtain citrus pulp elasticity difference data.

[0034] The present invention calculates the elastic modulus of the tactile perception data of different positions of individual citrus fruits, thereby obtaining the elastic modulus data of each part of the citrus fruits. This calculation provides a basis for evaluating the physical properties of fruits. The elastic modulus reflects the deformation ability and recovery ability of the fruit when subjected to force, and can reveal the texture differences of the fruit at different positions. Through this data, researchers can deeply understand the maturity and quality changes of various parts of the fruit. The surface temperature and humidity are extracted for the tactile perception data of different positions of individual citrus fruits to obtain the temperature and humidity data of each part. This step is intended to identify important environmental factors in the growth process of fruits, which directly affect the growth of citrus fruits and their intrinsic quality. By analyzing the surface temperature and humidity at different positions, researchers can understand the influence of temperature and humidity on the sugar and acid content and elastic properties of various parts of the fruit. Combining the elastic modulus data and the surface temperature and humidity data at different positions of the citrus fruits, elastic tension is calculated to obtain the elastic tension data of each part. This calculation provides a new perspective for quality evaluation of different parts of the fruit by combining physical properties with environmental factors. Elastic tension reflects the fruit's ability to respond when subjected to force, and can reveal the maturity state of each part and its taste characteristics. The sugar-acid ratio gradient data at different positions of individual citrus fruits are used to predict periodic deposition, and the periodic deposition data of the sugar-acid ratio are obtained. This process aims to identify the accumulation pattern of sugar and acid components inside the fruit by analyzing the changes in sugar and acid content in various parts of the fruit during different growth cycles through modeling. The difference in pulp elasticity is evaluated based on the periodic deposition data of the sugar-acid ratio of citrus and the elastic tension data at different positions, and the difference in pulp elasticity at various parts is obtained. The core of this evaluation is to comprehensively analyze the various indicators that affect the taste of the fruit and clarify the elastic characteristics of the pulp in different parts.

[0035] Preferably, the periodic deposition prediction of different citrus regions based on the citrus sugar-acid ratio gradient data comprises the following steps:

[0036] Based on the citrus sugar-acid ratio gradient data, the citrus cell osmotic pressure change data of different citrus regions were analyzed to obtain the citrus cell osmotic pressure change data;

[0037] According to the osmotic pressure change data of citrus cells, the cell wall expansion force simulation calculation of different citrus regions was performed to obtain the cell wall expansion force simulation data;

[0038] The cell membrane permeability data was obtained by analyzing the cell membrane permeability according to the osmotic pressure change data of citrus cells and the simulated data of cell wall swelling force;

[0039] Based on the citrus cell osmotic pressure change data and cell membrane permeability data, the periodic deposition of different citrus regions was predicted, and the periodic deposition data of the citrus sugar-acid ratio were obtained.

[0040] The present invention analyzes the citrus sugar-acid ratio gradient data, aiming to evaluate the changes in citrus cell osmotic pressure in different regions (i.e., different positions of individuals). Cell osmotic pressure is an important indicator that affects the balance of water and nutrients inside cells. By analyzing the changes in osmotic pressure in different parts in detail, researchers can reveal the physiological response of fruits under different growth conditions and further understand the maturity, quality and flavor characteristics of each part. This in-depth analysis not only provides data support for subsequent fruit quality evaluation, but also helps fruit farmers adjust management measures to improve the quality of fruits in specific areas. Based on the cell osmotic pressure change data obtained in the first step, researchers simulated and calculated the cell wall swelling force in different citrus regions. Cell wall swelling force is a key factor in determining cell growth and fruit expansion, and this simulation provides the necessary data basis for evaluating the mechanical properties of cells. By calculating the cell wall swelling force, researchers can gain a deep understanding of the growth potential and elasticity of various parts of the fruit. This data not only helps analyze the taste and texture of the fruit, but also provides fruit farmers with management suggestions on the fruit during its growth process. For example, by identifying differences in expansion force in different regions, fruit farmers can optimize fertilization and irrigation strategies to achieve more uniform fruit development and improve overall yield and quality. Combining cell osmotic pressure change data and cell wall expansion force simulation data, cell membrane permeability analysis is performed. The permeability of the cell membrane is directly related to the transport efficiency of water and nutrients, which in turn affects the growth and quality of the fruit. Through this analysis, researchers can evaluate the cell membrane characteristics of fruits in different parts and reveal the physiological state and maturity characteristics of each region. This understanding enables fruit farmers to accurately grasp the growth needs of different parts and formulate more effective management strategies. Good cell membrane permeability not only helps to improve the taste and sugar-acid ratio of the fruit, but also predicts the periodic deposition of different citrus regions based on cell osmotic pressure change data and cell membrane permeability data. The core of this process is to reveal the law of changes in the sugar-acid ratio of fruits in different growth cycles. By establishing a periodic deposition model, researchers can predict the future sugar and acid accumulation of fruits in various regions.

[0041] Preferably, step S3 comprises the following steps:

[0042] Step S31: analyzing the texture characteristics of citrus in different maturity areas on the internal maturity grade data of citrus according to the spectral characteristic band data to obtain maturity difference texture characteristic data;

[0043] Step S32: performing external morphological perception analysis of different maturity areas on the maturity difference texture feature data according to the tactile perception data of the citrus to obtain external morphological perception data of the citrus.

[0044] The present invention analyzes the internal maturity level of citrus according to the spectral characteristic band data, focusing on the texture characteristics of different maturity areas. This analysis uses spectral technology to accurately extract subtle texture changes inside citrus, thereby revealing the differences in structure and texture of fruits at different maturity stages. By comparing the texture characteristics of different regions, researchers can clearly identify which parts of the fruit show obvious texture changes during the ripening process. Based on the tactile perception data of citrus, the maturity difference texture feature data is subjected to external morphological perception analysis. The core of this step is to deeply explore the external morphological changes of fruits in different maturity areas through tactile perception means, including important indicators such as the surface smoothness, shape uniformity and hardness of the fruit. Through this comprehensive analysis, researchers can judge how the external morphology of the fruit changes with the change of internal maturity, thereby revealing the impact of maturity differences in different regions on the appearance quality.

[0045] Preferably, step S32 includes the following steps:

[0046] Step S321: analyzing the texture concavity and convexity of different maturity areas on the maturity difference texture feature data according to the tactile perception data of the citrus fruits, and obtaining the texture concavity and convexity data of different maturity areas;

[0047] Step S322: performing surface force transmission difference analysis on the texture concave-convex data of different maturity areas to obtain surface force transmission difference data;

[0048] Step S323: Calculating the surface compression force of citrus fruits at different maturity levels based on the surface force transmission difference data, and obtaining surface compression force data of citrus fruits at different maturity levels;

[0049] Step S324: performing surface force recovery and rebound analysis on the surface compression force data of different maturity areas to obtain surface force recovery and rebound data of different maturity areas;

[0050] Step S325: Perform external morphology perception analysis of different maturity areas based on the surface force recovery rebound data of different maturity areas and the texture concave-convex data of different maturity areas to obtain citrus external morphology perception data.

[0051] The present invention analyzes the texture concave-convex sense of different maturity areas based on the tactile perception data of citrus. This analysis can quantify the texture characteristics of the surface of citrus fruit through meticulous tactile measurement, revealing the degree of concave-convex in each maturity area. This data can not only reflect the maturity state of the fruit, but also show the external performance of the fruit after being affected by environmental factors during the growth process. By analyzing the concave-convex sense of different parts, researchers can identify the relationship between maturity and the texture of the fruit surface. For example, the surface of a mature fruit will show different concave-convex senses due to changes in the cell wall structure, thereby affecting the choice and preference of consumers. The surface force transmission difference analysis is performed on the texture concave-convex sense data of different maturity areas. This analysis aims to reveal how the fruit surface transmits force to the internal structure when it is subjected to external pressure. This difference in force transmission is not only related to the maturity of the fruit, but also affected by the texture of the fruit surface. By analyzing the characteristics of force transmission, researchers can better understand the response characteristics of the fruit under external stimuli, which is of great significance for evaluating the pressure resistance and vulnerability of the fruit. Based on the surface force transmission difference data, researchers calculate the surface compression force of citrus in different maturity areas. This calculation provides an important basis for evaluating the mechanical properties of fruits at different maturity stages, and can reveal the compressive resistance of fruits under external forces. By analyzing the compression force data, researchers can identify which areas show strong deformation ability when under pressure and which areas are more susceptible to damage. This understanding is crucial for fruit farmers' planting management and market sales strategies. The surface compression force data of different maturity areas are analyzed for surface force recovery and rebound. This analysis aims to evaluate whether the fruit can quickly recover to its original shape after compression, reflecting the elasticity and toughness of the fruit. The resilience of the fruit is directly related to its damage resistance during circulation, especially in market competition, it is crucial to maintain good appearance and quality. By analyzing the rebound data, researchers can identify the recovery characteristics of different maturity areas after compression, which not only helps fruit farmers optimize picking and transportation strategies, but also provides a scientific basis for subsequent quality control. Combining the surface force recovery and rebound data of different maturity areas and the texture concave and convex data, external morphological perception analysis is performed. The purpose of this analysis is to comprehensively evaluate the appearance characteristics and tactile perception performance of fruits at different maturity stages. By combining resilience with texture characteristics, researchers can more fully understand how the external morphology of fruits affects consumers' perception and purchasing decisions. This comprehensive analysis not only provides fruit farmers with specific suggestions for improving fruit quality, but also provides a basis for marketing and helps formulate more effective promotion strategies. Understanding the external morphological perceptual characteristics of different maturity areas will further enhance the market competitiveness of citrus.

[0052] Preferably, step S4 comprises the following steps:

[0053] Step S41: Based on the proximal strategy optimization algorithm, the external morphology perception data of the citrus and the internal maturity grade data of the citrus are subjected to citrus internal and external quality memory learning to obtain citrus internal and external quality memory learning data;

[0054] Step S42: normalizing the citrus internal and external quality memory learning data to obtain citrus internal and external quality memory normalized data;

[0055] Step S43: Design citrus internal and external quality selection firmware based on the normalized data of citrus internal and external quality memory, obtain citrus internal and external quality selection firmware, and send the citrus internal and external quality selection firmware to the cloud platform to perform citrus internal and external quality inspection and grading.

[0056] Based on the proximal strategy optimization algorithm, the present invention conducts a comprehensive analysis of the external morphological perception data and internal maturity grade data of citrus, and performs memory learning of the internal and external quality of citrus. This process enables the system to identify the relationship between different maturity areas and appearance characteristics, thereby forming a knowledge base that can accurately capture the quality characteristics of citrus. This memory learning of internal and external quality can not only improve the accuracy of fruit quality assessment, but also provide a scientific basis for subsequent selection and grading. By continuously optimizing the learning algorithm, the system can adapt to the characteristics of citrus under different varieties and different growth environments, thereby improving the accuracy of overall quality detection. The memory learning data of the internal and external quality of citrus is normalized to eliminate the dimensional differences between different data sources, so that various data can be compared and analyzed under the same standard. This process is a key step to ensure the accuracy of data processing and subsequent analysis. By standardizing various indicators, researchers can eliminate analytical deviations caused by different data scales and enhance the stability and robustness of the model. The normalized data provides a clearer and more consistent basis for subsequent firmware design and quality detection, so that the system can more accurately reflect the true characteristics of citrus in different regions when performing quality evaluation. This processing method not only improves the efficiency of data utilization, but also lays a solid data foundation for quality grading and market positioning. The internal and external quality selection firmware is designed based on the normalized data of the internal and external quality of citrus, and sent to the cloud platform to perform quality inspection and grading. This design combines machine learning and intelligent hardware to evaluate the quality of citrus in real time and realize an efficient and accurate automated grading process. The design of the firmware enables the system to intelligently judge the quality of citrus in different regions based on the existing data model, and effectively classify and process it. This process not only improves the detection efficiency and reduces the need for manual intervention, but also enhances the accuracy of quality control. At the same time, the application of the cloud platform enables data to be updated and stored in a timely manner, realizes centralized management and analysis of data, and provides strong support for subsequent market decisions. Through this efficient quality inspection and grading method.

[0057] Preferably, the present invention further provides a citrus internal and external quality detection and grading system based on deep learning, which is used to execute the citrus internal and external quality detection and grading method based on deep learning as described above, and the citrus internal and external quality detection and grading system based on deep learning comprises:

[0058] The data acquisition module is used to collect spectral images of citrus through a hyperspectral imager to obtain spectral characteristic band data; collect tactile perception data of citrus through a tactile sensor built into the mechanical claw to obtain tactile perception data of citrus;

[0059] The internal maturity grade classification module is used to identify the sugar-acid ratio gradient of different citrus regions according to the spectral characteristic band data, and obtain the citrus sugar-acid ratio gradient data; evaluate the difference in pulp elasticity of citrus in different regions according to the citrus sugar-acid ratio gradient data and the citrus tactile perception data, and obtain the citrus pulp elasticity difference data; divide the internal maturity grades of different citrus regions based on the citrus sugar-acid ratio gradient data and the citrus pulp elasticity prediction data, and obtain the citrus internal maturity grade data;

[0060] The external morphology perception module is used to perform external morphology perception analysis on the internal maturity grade data of citrus fruits in different maturity areas according to the spectral characteristic band data to obtain external morphology perception data of citrus fruits;

[0061] The firmware design module is used to learn the internal and external quality of citrus by memorizing the external morphological perception data and the internal maturity grade data of citrus based on the proximal strategy optimization algorithm, and memorize the normalized data of the internal and external quality of citrus; the firmware for selecting the internal and external quality of citrus is designed based on the normalized data of the internal and external quality of citrus, and the firmware for selecting the internal and external quality of citrus is obtained, and the firmware for selecting the internal and external quality of citrus is sent to the cloud platform to perform internal and external quality detection and grading of citrus.

[0062] The beneficial effect of the present invention is that the accuracy and comprehensiveness of quality inspection are greatly improved by collecting spectral images and tactile perception data of citrus through a hyperspectral imager and a tactile sensor. The hyperspectral imager can obtain detailed spectral features of the surface of citrus, and these spectral data can reveal the chemical composition and state of the fruit, especially the sugar and acid content. This process can effectively identify the characteristics of the fruit at different stages of maturity through the analysis of different bands. In addition, the built-in tactile sensor obtains the hardness and elasticity data of citrus through contact, supplementing the physical properties that spectral imaging cannot fully reflect. This combination of multidimensional data provides a solid foundation for subsequent classification, so that the quality inspection of citrus is not limited to visual features, but comprehensively considers the chemical composition and physical properties, so that the maturity and flavor characteristics of the fruit can be more accurately evaluated. The sugar-acid ratio gradient of different regions of citrus is identified using spectral feature band data, which is an important basis for quality classification. The sugar-acid ratio is an important indicator for evaluating the taste of fruit, which directly affects consumer choice and market acceptance. By performing gradient identification on the ratios of different regions, it is possible to identify which regions of the fruit are sweeter or more sour, and thus classify them into different consumer markets or usage purposes. In addition, the tactile perception data of citrus is combined to evaluate the difference in pulp elasticity, providing a more comprehensive perspective on the internal quality of the fruit. Fruits with better elasticity usually mean higher water content and better taste, which is crucial for marketing. The comprehensive evaluation at this stage makes the classification of citrus more scientific and provides data support for subsequent sales strategies. By performing external morphological perception analysis on the internal maturity grade data of citrus, a more comprehensive understanding of the relationship between the apparent characteristics and the intrinsic quality of the fruit can be achieved. Using spectral feature band data, this process not only focuses on the physical characteristics of the fruit, such as color, size, and shape, but also infers the internal maturity through these external morphological characteristics. This method allows researchers and fruit farmers to systematically analyze citrus in different maturity areas, so as to formulate corresponding picking and sales strategies. Through external morphological perception data, the health status, maturity, and suitable market positioning of the fruit can be quickly identified, which improves the efficiency and accuracy of classification. In addition, this step provides a scientific basis for subsequent quality evaluation and selection, ensuring that consumers can obtain better quality citrus products, thereby improving the efficiency and benefits of the entire supply chain. The process of memorizing and learning the internal and external quality data of citrus based on the proximal strategy optimization algorithm marks a major advancement in fruit quality detection and grading. By combining external morphological perception data with internal maturity grade data, the algorithm can identify the key factors affecting citrus quality and form an efficient quality memory model. This model can normalize different data so that each feature has consistent standards for subsequent analysis. Based on these normalized data, the designed citrus internal and external quality selection firmware can intelligently identify and classify fruits of different qualities, further improving the automation and intelligence level of detection.This method not only improves the detection efficiency, but also reduces the interference of human factors, ensuring that each citrus can be evaluated scientifically and objectively. By combining the selection firmware with the cloud platform, data can be uploaded and analyzed in real time, providing a powerful data support and feedback mechanism for the entire industry, and ultimately promoting the intelligent and modern process of citrus quality management. Therefore, the present invention provides an optimization process for a traditional method for detecting and grading the internal and external quality of citrus based on deep learning, solves the problems of inaccurate recognition of the internal and external maturity of citrus and low degree of automation in a traditional method for detecting and grading the internal and external quality of citrus based on deep learning, improves the accuracy of identifying the internal and external maturity of citrus, and improves the level of automation. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a flowchart of the steps of a method for detecting and grading the internal and external quality of citrus based on deep learning;

[0064] Figure 2 for Figure 1 Detailed implementation steps of step S2 in FIG.

[0065] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0066] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0067] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0068] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0069] To achieve this, please refer to Figure 1 to Figure 2 A method for detecting and grading the internal and external quality of citrus based on deep learning, the method comprising the following steps:

[0070] Step S1: collecting spectral images of citrus through a hyperspectral imager to obtain spectral characteristic band data; collecting tactile perception data of citrus through a tactile sensor built into a mechanical claw to obtain tactile perception data of citrus;

[0071] Step S2: identifying the sugar-acid ratio gradient of different regions of citrus according to the spectral characteristic band data, and obtaining the sugar-acid ratio gradient data of citrus; evaluating the difference in pulp elasticity of citrus in different regions according to the sugar-acid ratio gradient data of citrus and the tactile perception data of citrus, and obtaining the difference in pulp elasticity of citrus; dividing the internal maturity level of different regions of citrus based on the sugar-acid ratio gradient data of citrus and the citrus pulp elasticity prediction data, and obtaining the internal maturity level data of citrus;

[0072] Step S3: performing external morphological perception analysis of different maturity areas on the internal maturity grade data of the citrus according to the spectral characteristic band data to obtain external morphological perception data of the citrus;

[0073] Step S4: Based on the proximal strategy optimization algorithm, the external morphology perception data of the citrus and the internal maturity grade data of the citrus are memorized and learned to obtain the normalized internal and external quality data of the citrus; based on the normalized internal and external quality memory data of the citrus, the internal and external quality selection firmware of the citrus is designed to obtain the internal and external quality selection firmware of the citrus, and the internal and external quality selection firmware of the citrus is sent to the cloud platform to execute the internal and external quality detection and grading of the citrus.

[0074] In the embodiment of the present invention, reference Figure 1 The above is a schematic flow chart of the steps of a method for detecting and grading the internal and external quality of citrus based on deep learning of the present invention. In this example, the method for detecting and grading the internal and external quality of citrus based on deep learning includes the following steps:

[0075] Step S1: collecting spectral images of citrus through a hyperspectral imager to obtain spectral characteristic band data; collecting tactile perception data of citrus through a tactile sensor built into a mechanical claw to obtain tactile perception data of citrus;

[0076] In an embodiment of the present invention, a spectral image of citrus is collected by a hyperspectral imager to obtain spectral reflection information of different areas on the surface of citrus. The hyperspectral imager controls a light source of a specific band to illuminate the citrus, uses a sensor to record the reflected light on the surface of the citrus in real time, decomposes light of different wavelengths through an optical spectrometer, and forms multi-dimensional spectral image data. In specific implementation, the instrument continuously scans the surface of the citrus to ensure that each area is evenly covered, and at the same time, the original spectral image is converted into analyzable spectral characteristic band data through a spectral image processing algorithm. The data contains the spectral characteristics of different areas of citrus in a specific band, which can reflect the internal chemical composition and structure of different areas. In the tactile perception data collection, a high-precision tactile sensor built into the mechanical claw is used, and the surface of the citrus is multi-touch sensed through the robotic arm automated operating system to obtain tactile information such as surface hardness, softness and elasticity. Based on the piezoelectric effect or capacitive sensing principle, the sensor converts the tiny deformation of the citrus surface during the contact process into an electrical signal, and performs high-frequency sampling to form citrus tactile perception data. This data is used to characterize the physical properties of different parts of citrus, especially the differences in hardness and softness of the skin, to support subsequent steps.

[0077] Step S2: identifying the sugar-acid ratio gradient of different regions of citrus according to the spectral characteristic band data, and obtaining the sugar-acid ratio gradient data of citrus; evaluating the difference in pulp elasticity of citrus in different regions according to the sugar-acid ratio gradient data of citrus and the tactile perception data of citrus, and obtaining the difference in pulp elasticity of citrus; dividing the internal maturity level of different regions of citrus based on the sugar-acid ratio gradient data of citrus and the citrus pulp elasticity prediction data, and obtaining the internal maturity level data of citrus;

[0078] In an embodiment of the present invention, the sugar-acid ratio of citrus is identified by a chemometric method according to the spectral characteristic band data obtained in step S1. In a specific operation, the spectral characteristics are interpreted by partial least squares (PLS) or principal component regression (PCR), and a mathematical model is established between the spectral band data and the concentration changes of sugars and acids in citrus to analyze the sugar-acid ratio gradients in different regions. The method calculates the sugar-acid ratio of each region by performing multiple linear regression on the spectral reflection intensity of different bands, and finally obtains the citrus sugar-acid ratio gradient data, which reflects the distribution of chemical components inside the citrus. Based on the citrus sugar-acid ratio gradient data and the tactile perception data obtained in step S1, the elasticity difference of the pulp in different regions of citrus is evaluated. During the operation, an elastic modulus calculation method is used, and the deformation amount generated after the citrus surface is subjected to pressure is calculated according to the force-displacement curve of the tactile perception data, and the elastic coefficient of different regions is obtained. Combined with the sugar-acid ratio gradient data, linear regression analysis is used to determine the elasticity difference of the pulp in different regions of citrus. Through this evaluation, the flexibility and hardness distribution of citrus pulp can be determined, and the citrus pulp elasticity difference data can be obtained. Based on the citrus sugar-acid ratio gradient data and the citrus pulp elasticity difference data, the internal maturity levels of different citrus regions are divided. First, the sugar-acid ratio and elasticity data are multi-dimensionally classified using the K-means clustering algorithm or the support vector machine classification method. In the specific operation, by calculating the sugar-acid ratio and elasticity data of each region, a multidimensional data space is constructed to divide the various regions of citrus into several categories. Then, based on these categories, the maturity of different regions is divided into multiple levels, such as "immature", "semi-mature", "mature" and "overripe". Through these data analysis, the internal maturity level data of citrus is finally obtained. This data can help further judge the overall maturity state of citrus and provide a basis for subsequent quality grading.

[0079] Step S3: performing external morphological perception analysis of different maturity areas on the internal maturity grade data of the citrus according to the spectral characteristic band data to obtain external morphological perception data of the citrus;

[0080] In the embodiment of the present invention, the external morphology of citrus is first analyzed using the spectral feature band data collected in step S1, the main purpose of which is to associate different maturity areas of citrus with external morphological features. In the specific operation, the external morphology of citrus is extracted and analyzed by using edge detection and morphological analysis methods (such as edge enhancement algorithms based on Laplace operators and morphological opening and closing operations). The specific bands of the spectral image reflect the physical structural differences in different areas of citrus. Combined with these data, the system can identify the external skin texture, concave area, glossiness and color distribution of citrus. In order to improve the recognition accuracy, Fourier transform is used for frequency domain analysis to determine the subtle ripple changes of the skin. Then, these external morphological features are associated with the internal maturity grade data of citrus obtained in step S2. Specifically, the mapping relationship between the external morphology of citrus and the internal maturity is established by the multivariate linear regression method, so as to analyze the external morphological features corresponding to different maturity areas. The result of this analysis is to obtain the external morphological perception data of citrus, which reflects the relationship between the specific surface features of citrus and the internal maturity, and provides support for subsequent steps.

[0081] Step S4: Based on the proximal strategy optimization algorithm, the external morphology perception data of the citrus and the internal maturity grade data of the citrus are memorized and learned to obtain the normalized internal and external quality data of the citrus; based on the normalized internal and external quality memory data of the citrus, the internal and external quality selection firmware of the citrus is designed to obtain the internal and external quality selection firmware of the citrus, and the internal and external quality selection firmware of the citrus is sent to the cloud platform to execute the internal and external quality detection and grading of the citrus.

[0082] In an embodiment of the present invention, based on the external morphology perception data of citrus obtained in step S3 and the internal maturity grade data of citrus in step S2, the proximal policy optimization (PPO) algorithm is used to perform memory learning of the internal and external qualities of citrus. Proximal policy optimization is a reinforcement learning algorithm, which is particularly suitable for dynamic learning of such multi-dimensional data. In specific implementation, by introducing a multi-dimensional data space, the external morphology and internal maturity of citrus are used as input features. In each iteration, the algorithm samples and evaluates different internal and external quality combinations, and updates the strategy function according to the obtained reward value. In this process, the parameters of the strategy function are continuously optimized to achieve the best matching degree of the internal and external qualities of citrus. Through this reinforcement learning process, the system will gradually memorize the correlation between the internal and external qualities of citrus, and normalize this learning result to obtain the normalized data of the internal and external qualities of citrus. The purpose of normalization is to map multi-dimensional data into a unified standard range for subsequent analysis and processing. Based on the normalized data of the internal and external qualities of citrus, a firmware for selecting the internal and external qualities of citrus is designed. The core of the firmware design is to determine the execution logic of the automated detection and grading system. The firmware mainly sets different quality classification standards by threshold judgment on normalized data. In specific operations, the threshold setting rules are used to grade citrus according to multiple dimensions such as maturity, surface morphology, and pulp elasticity. Through the decision tree algorithm, classification paths for practical application are designed. The firmware will automatically screen and grade citrus according to these paths during operation. Finally, the designed citrus internal and external quality selection firmware is uploaded to the cloud platform. The cloud platform has the ability to quickly deploy and execute, and can perform real-time quality detection and grading operations on large-scale citrus. The firmware completes the detection and final grading of the internal and external quality of citrus through calculation and execution on the cloud platform, ensuring efficiency and accuracy in large-scale agricultural production.

[0083] Preferably, step S1 comprises the following steps:

[0084] Step S11: collecting spectral images of citrus using a hyperspectral imager to obtain spectral images of citrus;

[0085] Step S12: performing image enhancement processing on the citrus spectrum image to obtain a citrus spectrum enhanced image;

[0086] Step S13: extracting characteristic bands from the citrus spectrum enhanced image to obtain spectral characteristic band data;

[0087] Step S14: collecting tactile perception data of the citrus through the tactile sensor built into the mechanical claw to obtain tactile perception data of the citrus.

[0088] In an embodiment of the present invention, a hyperspectral imager is used to collect spectral images of the citrus surface in all directions and at multiple angles. The hyperspectral imager records the spectral information reflected by the citrus surface by scanning the surface, covering multiple bands from visible light to near infrared. In order to ensure the integrity and accuracy of data acquisition, the light source of the imager must be in a stable state to avoid interference from external ambient light. The imager generates a three-dimensional data cube in a point-by-point scanning manner, which contains the spectral information of each pixel. After preliminary acquisition, the generated citrus spectral image is stored in a spectral data matrix, each row of which represents a different spectral band, and each column corresponds to the spectral reflection intensity at different positions on the citrus surface. The citrus spectral image obtained from step S11 is subjected to image enhancement processing. The purpose of image enhancement is to highlight the specific spectral features of the citrus surface to improve the accuracy of subsequent analysis. In this process, the brightness and contrast of the spectral image are first adjusted using the histogram equalization technique, so that the spectral information of the dark area of ​​the citrus surface is effectively enhanced. Next, an adaptive filter is applied to remove noise in the image to retain important information in the spectral data. In addition, the edge enhancement of the image is performed by the Laplace sharpening algorithm to ensure that the spectral information of the detailed part of the citrus surface is clearer. The result of this series of operations is an enhanced citrus spectral image, which provides clearer spectral data for the subsequent extraction of band features. The spectral feature band data is extracted from the enhanced spectral image using band selection and feature extraction algorithms. This operation first reduces the dimension of the spectral data through principal component analysis (PCA) to extract the main components in the spectral information, thereby eliminating redundant bands and noise interference. The spectral data after dimension reduction is further analyzed, and the spectral response curve is used to identify the reflectance characteristics of citrus in a specific wavelength range, especially the bands related to quality indicators such as sugar-acid ratio, pigment content, and skin thickness. In this process, the maximum entropy band selection method is used for band optimization to ensure that the extracted spectral bands are highly correlated with the quality of citrus. After these processes, the spectral feature band data obtained is used for subsequent quality analysis and detection. The tactile perception data of citrus is collected by the high-precision tactile sensor built into the mechanical claw. The robotic claw first gently touches the surface of the citrus in a precise force control mode to ensure that the peel is not damaged. The sensor detects the mechanical response of the citrus surface in real time and records the tactile information fed back during the touch process, such as hardness, elasticity, compression deformation, etc. In order to obtain comprehensive tactile perception data, the robotic claw performs multiple tactile sampling at different contact points and different pressures. The collected tactile data includes indicators such as the degree of deformation and elastic recovery time of different areas of citrus under specific pressure. These tactile data can reflect the differences in the pulp structure and skin thickness of citrus, providing a basis for quality analysis and internal maturity assessment in subsequent steps.

[0089] Preferably, step S2 comprises the following steps:

[0090] Step S21: using a preset spectral band similarity recognition model to classify the spectral feature band data into similar bands to obtain spectral feature band similarity data;

[0091] Step S22: identifying the sugar-acid ratio gradient of different regions of citrus according to the similarity data of the spectral characteristic bands, and obtaining the sugar-acid ratio gradient data of citrus;

[0092] Step S23: evaluating the difference in pulp elasticity of citrus in different areas according to the citrus sugar-acid ratio gradient data and the citrus tactile perception data, and obtaining citrus pulp elasticity difference data;

[0093] Step S24: Based on the citrus sugar-acid ratio gradient data and the citrus pulp elasticity difference data, the internal maturity grade of different citrus areas is divided to obtain the citrus internal maturity grade data.

[0094] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes:

[0095] Step S21: using a preset spectral band similarity recognition model to classify the spectral feature band data into similar bands to obtain spectral feature band similarity data;

[0096] In an embodiment of the present invention, a spectral band similarity recognition algorithm is used to classify similar bands of the spectral feature band data extracted from step S13. First, a spectral similarity matrix is ​​constructed based on the correlation and similarity calculation formula between spectral bands. The matrix is ​​numerically analyzed based on the reflectance characteristics of citrus in different bands, and the Pearson correlation coefficient is usually used to quantify the linear relationship between the bands. Next, a hierarchical clustering algorithm is used to classify bands with highly similar reflectance characteristics into the same category, thereby generating spectral feature band similarity data. This process ensures that the classification results have high sensitivity and can minimize the redundancy of band information through threshold setting. Ultimately, the obtained spectral feature band similarity data will provide a basis for subsequent sugar-acid ratio gradient identification.

[0097] Step S22: identifying the sugar-acid ratio gradient of different regions of citrus according to the similarity data of the spectral characteristic bands, and obtaining the sugar-acid ratio gradient data of citrus;

[0098] In an embodiment of the present invention, based on the similar data of the spectral characteristic bands generated in step S21, the sugar-acid ratio gradients of different regions of citrus are identified using a multiple regression analysis method. First, the reflection characteristics in the similar bands are fitted with the known sugar-acid ratio data in the citrus fruit, and the regression coefficients are optimized by the least squares method to establish a mapping relationship between the sugar-acid ratio of the citrus region and the spectral bands. This process identifies the variation pattern of the sugar-acid ratio between the citrus skin and the pulp through piecewise linear regression, and forms a gradient distribution map. The obtained citrus sugar-acid ratio gradient data reflects the difference in the sugar-acid ratio of each region from the center to the edge of the citrus, providing a scientific basis for the subsequent elastic difference evaluation.

[0099] Step S23: evaluating the difference in pulp elasticity of citrus in different areas according to the citrus sugar-acid ratio gradient data and the citrus tactile perception data, and obtaining citrus pulp elasticity difference data;

[0100] In an embodiment of the present invention, based on the citrus sugar-acid ratio gradient data obtained in step S22, and in combination with the citrus tactile perception data collected in step S14, an assessment is made of the difference in elasticity of the citrus pulp in different regions. This step quantifies the difference in elasticity of the pulp in different regions through the elastic modulus calculation formula, combined with the change in the sugar-acid ratio. Specifically, the mechanical response value in the citrus tactile perception data is used, combined with the distribution of the sugar-acid ratio in different regions, and the internal stress distribution of the citrus is simulated by the finite element analysis method, and then the elastic modulus of the pulp in each region is calculated. The resulting citrus pulp elasticity difference data reflects the changes in the hardness and softness of the pulp, providing a mechanical basis for the final maturity assessment.

[0101] Step S24: Based on the citrus sugar-acid ratio gradient data and the citrus pulp elasticity difference data, the internal maturity grade of different citrus areas is divided to obtain the citrus internal maturity grade data.

[0102] In an embodiment of the present invention, based on the citrus sugar-acid ratio gradient data obtained in step S22 and the citrus pulp elasticity difference data obtained in step S23, different regions of citrus are divided into internal maturity levels. The operation first assigns weights to the sugar-acid ratio gradient data and the pulp elasticity difference data through the analytic hierarchy process (AHP). Then, the comprehensive maturity of different regions is calculated based on the weighted average method to form a plurality of grade intervals. To ensure the scientificity and accuracy of the classification, the fuzzy clustering method is used to divide different regions of citrus into a plurality of maturity levels, and a corresponding regional maturity map is generated. The citrus internal maturity grade data finally obtained provides a clear basis for fruit grading, ensuring that each citrus can be accurately classified according to its internal quality.

[0103] Preferably, step S22 includes the following steps:

[0104] Step S221: extracting sugar and acid bands of different regions of citrus from similar data of spectral characteristic bands to obtain sugar and acid spectral band data;

[0105] Step S222: performing multi-region sugar-acid distribution chromatography processing on different regions of citrus based on the sugar-acid spectral band data to obtain multi-region sugar-acid distribution chromatography data;

[0106] Step S223: performing band density clustering calculation on the sugar-acid spectrum band data of different regions of citrus according to the multi-region sugar-acid distribution chromatography data to obtain sugar-acid distribution band density clustering data;

[0107] Step S224: identifying the sugar-acid ratio gradients of different citrus regions using the sugar-acid distribution band density clustering data to obtain citrus sugar-acid ratio gradient data.

[0108] In an embodiment of the present invention, for the spectral characteristic band similarity data obtained in step S21, the spectral band characteristic analysis method is used to extract the sugar and acid bands of different regions of citrus. First, the spectral characteristic bands of citrus are segmented, and the complex band data are converted into spectrum data by Fourier transform using the variation characteristics of the spectral reflectance coefficient, and the spectral bands related to the sugar and acid content of citrus are screened out. Next, the sugar and acid content is fitted with the spectral bands by multivariate correction technology, such as partial least squares (PLS), to ensure that the selected band data can accurately reflect the sugar and acid content. The sugar and acid spectral band data finally extracted contain the sugar and acid content variation information of citrus fruits in different regions, providing a basis for subsequent sugar and acid distribution chromatography. Based on the sugar and acid spectral band data generated in step S221, the sugar and acid distribution of different regions of citrus is processed in multiple regions using tomography technology. First, the principle of spectral tomography is applied, and the sugar and acid content information of different depths inside the citrus is obtained layer by layer by tomography of the spectral bands, forming multiple spectral slices. Through the three-dimensional reconstruction algorithm, the data of each slice is superimposed to form the sugar-acid distribution tomographic image of citrus. In order to further refine the tomographic data, a three-dimensional interpolation algorithm based on voxels (volumetric pixel) is used to smooth the sugar-acid distribution between different layers to ensure that the final multi-region sugar-acid distribution tomographic data can accurately reflect the sugar-acid content differences of citrus fruits at all levels. According to the multi-region sugar-acid distribution tomographic data generated in step S222, the density clustering algorithm (DBSCAN) is used to perform band density clustering calculation on the sugar-acid spectral band data. First, by calculating the density change in the sugar-acid distribution band, the neighborhood radius parameter is set to identify the high-density and low-density areas of sugar-acid distribution in different regions of citrus. The density clustering algorithm automatically classifies the bands in different regions based on the distribution characteristics of sugar-acid content. By calculating the frequency of occurrence of each band in a specific area, a cluster center and a corresponding density distribution map are formed. The sugar-acid distribution band density clustering data finally generated can reflect the sugar-acid distribution law of citrus in different regions, and provide a classification basis for subsequent gradient recognition. Based on the sugar-acid distribution band density clustering data generated in step S223, the gradient analysis method is used to perform gradient identification on the sugar-acid ratios of different areas of citrus. This step first calculates the sugar-acid ratios of each clustering area and uses the gradient field model to quantify the spatial variation of sugar-acid content. The gradient field model calculates the rate and direction of change of the sugar-acid ratio of citrus fruit in different areas by constructing local derivatives, and clarifies the high and low distribution trends of sugar-acid content in each area. Next, the continuous gradient interpolation technology is used to smooth the sugar-acid ratios in the edge areas to ensure that the identified gradient distribution has high accuracy. The final generated citrus sugar-acid ratio gradient data provides an important numerical basis for the grading and detection of the internal quality of citrus, ensuring the accuracy and consistency in the citrus quality detection process.

[0109] Preferably, step S23 includes the following steps:

[0110] Step S231: calculating the elastic modulus of different regions of the citrus based on the tactile perception data of the citrus to obtain regional elastic modulus data of the citrus;

[0111] Step S232: extracting the surface temperature and humidity of different areas of the citrus from the tactile perception data of the citrus to obtain the surface temperature and humidity data of the citrus area;

[0112] Step S233: Calculate the elastic tension of different citrus regions according to the elastic modulus data of the citrus regions and the surface temperature and humidity data of the citrus regions to obtain the elastic tension data of the citrus regions;

[0113] Step S234: performing periodic deposition prediction of different citrus regions on the citrus sugar-acid ratio gradient data to obtain citrus sugar-acid ratio periodic deposition data;

[0114] Step S235: Evaluate the difference in pulp elasticity of different regions of citrus according to the citrus sugar-acid ratio periodic deposition data and the citrus regional elastic tension data to obtain citrus pulp elasticity difference data.

[0115] In the embodiment of the present invention, for the citrus tactile perception data obtained in step S14, the elastic modulus calculation method in material mechanics is used to quantitatively analyze the elastic modulus of different regions of the citrus. The elastic modulus of the region is calculated by measuring the deformation of the citrus in different regions under the action of external forces using the force-displacement curve collected by the tactile sensor and the elastic mechanics formula, combined with the physical properties of the citrus pulp, by measuring the deformation of the citrus in different regions under the action of external forces. The specific operation process includes filtering the tactile perception data, removing noise interference, and using the slope of the stress-strain curve as the basis for estimating the elastic modulus. The citrus regional elastic modulus data finally generated provides basic parameters for subsequent elastic tension calculation. Based on the tactile perception data collected by the tactile sensor, the surface temperature and humidity of different regions of the citrus are extracted using a real-time feedback system embedded in a temperature and humidity sensor. First, local sampling is performed using a temperature and humidity sensor to record the temperature and humidity changes in each region of the citrus surface. In order to ensure the accuracy of the data, the temperature and humidity data collected by the sensor are denoised using a time series analysis method to eliminate the influence of environmental factors. Next, the temperature and humidity data of each region are normalized to ensure that the temperature and humidity differences between different regions can be quantified and compared. The surface temperature and humidity data of the citrus region finally generated reflects the physical state of each area on the surface of the citrus, and provides the environmental parameters required for the analysis of the elastic difference of the pulp. Combined with the elastic modulus data of the citrus region generated in step S231 and the surface temperature and humidity data of the citrus region in step S232, the elastic tension of different citrus regions is calculated. First, through the coupling relationship between the elastic modulus and the temperature and humidity, the thermal stress equation in thermodynamics is used, and then, based on the temperature and humidity state of each region, the elastic deformation and local tension of the pulp in different regions are calculated to obtain the elastic tension data of each region. The finite element analysis (FEA) method is used in this calculation process to numerically simulate and solve the elastic force field in different regions. The elastic tension data of the citrus region finally generated can accurately reflect the internal tension difference of the citrus pulp in different regions. Based on the citrus sugar-acid ratio gradient data generated in step S224, the sugar-acid ratio of different regions of citrus is periodically predicted by using time series data processing technology. First, by performing a partition analysis on the citrus sugar-acid ratio gradient data, the periodic law of the sugar-acid ratio changing over time is identified. Next, a periodic analysis algorithm (such as Fourier transform or wavelet analysis) is applied to extract the main periodic characteristics of the sugar-acid ratio, and then the deposition rate and trend of sugar-acid in different regions are predicted. In order to improve the accuracy of the prediction, the empirical mode decomposition (EMD) method is used to decompose the sugar-acid ratio time series to identify potential deposition cycles. The resulting citrus sugar-acid ratio periodic deposition data provides a temporal dimension for the evolution of the internal quality of citrus. The citrus sugar-acid ratio periodic deposition data generated in step S234 is combined with the citrus regional elastic tension data in step S233 to evaluate the difference in pulp elasticity in different citrus regions.First, the elastic tension of each region is coupled with the sugar-acid ratio deposition data. By analyzing the effect of the periodic changes of sugar and acid on the elasticity of the pulp, the elastic differences of the pulp in different regions are evaluated. This process uses multivariate regression analysis, takes the periodic deposition of the sugar-acid ratio and the elastic tension as input variables, and quantifies the elastic changes of the pulp through the stress-strain relationship model. The resulting citrus pulp elasticity difference data can accurately reflect the changes in the physical state of the pulp in different regions of citrus, providing an accurate basis for the internal quality detection of citrus.

[0116] Preferably, the periodic deposition prediction of different citrus regions based on the citrus sugar-acid ratio gradient data comprises the following steps:

[0117] Based on the citrus sugar-acid ratio gradient data, the citrus cell osmotic pressure change data of different citrus regions were analyzed to obtain the citrus cell osmotic pressure change data;

[0118] According to the osmotic pressure change data of citrus cells, the cell wall expansion force simulation calculation of different citrus regions was performed to obtain the cell wall expansion force simulation data;

[0119] The cell membrane permeability data was obtained by analyzing the cell membrane permeability according to the osmotic pressure change data of citrus cells and the simulated data of cell wall swelling force;

[0120] Based on the citrus cell osmotic pressure change data and cell membrane permeability data, the periodic deposition of different citrus regions was predicted, and the periodic deposition data of the citrus sugar-acid ratio were obtained.

[0121] In an embodiment of the present invention, based on the gradient data of the sugar-acid ratio of citrus, the osmotic pressure change of citrus cells in different regions is analyzed. First, according to the gradient data of the sugar-acid ratio of different citrus regions, the sugar-acid concentration inside the citrus cells is associated with the sugar-acid ratio of the external environment, and the osmotic pressure change of cells in each region is analyzed. The change in osmotic pressure mainly reflects the cell's ability to absorb and discharge water and nutrients. This step of analysis adopts a partitioning processing method, analyzes each region independently, and applies a gradient analysis algorithm in the data processing stage to capture the effect of the intercellular sugar-acid ratio gradient on the osmotic pressure change. The osmotic pressure change data of citrus cells finally output provides the necessary physical parameters for the subsequent calculation of cell wall swelling force. According to the osmotic pressure change data of citrus cells, the cell wall swelling force of different regions is simulated and calculated. First, the cell wall swelling force is directly affected by the change in osmotic pressure. Therefore, on the basis of the analysis of the change in osmotic pressure, the finite element analysis method is used to simulate the swelling force. In the specific calculation process, the mathematical relationship between the swelling force and the osmotic pressure change is established by simulating the deformation and stress of the cell wall when the osmotic pressure changes. The final generated cell wall swelling force simulation data provides a physical basis for the subsequent cell membrane permeability analysis. Based on the citrus cell osmotic pressure change data and cell wall swelling force simulation data, the cell membrane permeability of different citrus regions was analyzed. Cell membrane permeability is an important factor that determines the entry and exit of water and solutes into and out of cells. On the basis of osmotic pressure and swelling force, the permeability changes of cell membranes in different regions under different osmotic pressure and swelling force conditions were simulated, and the effects of membrane protein channels and osmotic pressure gradients on the flow of water and sugar and acid were combined, and the membrane permeability model in physical chemistry was used for calculation. Molecular dynamics simulation technology is also used in the cell membrane permeability analysis to accurately describe the permeability characteristics of cell membranes in different citrus regions. The final generated cell membrane permeability data reflects the material transport characteristics of citrus cells in different regions. According to the citrus cell osmotic pressure change data and cell membrane permeability data, periodic deposition predictions are made for different citrus regions. First, the periodic deposition law of sugar and acid substances in cells in different regions is determined by combining the changes in cell membrane permeability and osmotic pressure. Time series analysis techniques (such as Fourier transform or wavelet analysis) are used to extract periodic features of deposition data to predict the deposition cycle of sugar and acid. The periodic deposition prediction not only considers the spatiotemporal distribution of sugar and acid, but also simulates the accumulation process of sugar and acid in each region through the dynamic changes of cell membrane permeability and osmotic pressure. The resulting periodic deposition data of citrus sugar and acid ratios provide a scientific basis for the time dimension of citrus quality detection, and also reflect the distribution and evolution of sugar and acid in different regions of citrus.

[0122] Preferably, step S3 comprises the following steps:

[0123] Step S31: analyzing the texture characteristics of citrus in different maturity areas on the internal maturity grade data of citrus according to the spectral characteristic band data to obtain maturity difference texture characteristic data;

[0124] Step S32: performing external morphological perception analysis of different maturity areas on the maturity difference texture feature data according to the tactile perception data of the citrus to obtain external morphological perception data of the citrus.

[0125] In the embodiment of the present invention, according to the spectral characteristic band data, the citrus texture feature analysis of different maturity areas is performed on the internal maturity grade data of citrus, aiming to reveal the maturity differences of various areas inside the citrus. First, the spectral characteristic band data covering the visible light and near-infrared regions are obtained by using hyperspectral imaging technology, which can reflect the light absorption and reflection characteristics of citrus pulp. Then, the key band information related to the maturity of citrus is determined by the relationship between the spectral reflectance and the specific band, and a spectral characteristic band matrix is ​​formed. Next, the spectral data of different regions are subjected to principal component analysis (PCA) to extract the texture features related to the internal organizational structure and maturity of citrus. The texture analysis adopts the gray level co-occurrence matrix (GLCM) method to calculate the texture parameters (such as contrast, entropy, correlation, etc.) of the citrus in different maturity areas, so as to capture the subtle changes of maturity differences. The maturity difference texture feature data generated by this process provides important internal structure information support for the subsequent external morphological perception analysis. According to the tactile perception data of citrus, the maturity difference texture feature data is subjected to external morphological perception analysis of different maturity areas. First, collect tactile perception data on the surface of citrus. The tactile sensor can accurately capture the tiny morphological changes and physical properties of the fruit surface, such as parameters such as surface roughness, elasticity and hardness. These data are obtained through a multi-point tactile perception device, which can quantify the tactile characteristics of the outer surface of the citrus. Then, these external morphological features are cross-analyzed with the maturity difference texture feature data generated in step S31. Using the differences in texture data, combined with the tactile perception results, edge detection and morphological operation technology are used to extract the external morphological features of different maturity areas. Through Fourier transform and shape descriptor algorithms (such as Fourier coefficients, curvature, etc.), multi-scale analysis of the surface morphology of citrus is performed to evaluate the appearance differences of fruits at different maturity stages. The final generated citrus external morphological perception data can accurately reflect the impact of maturity on the appearance of citrus, and provide a reliable basis for the comprehensive evaluation and grading of citrus quality.

[0126] Preferably, step S32 includes the following steps:

[0127] Step S321: analyzing the texture concavity and convexity of different maturity areas on the maturity difference texture feature data according to the tactile perception data of the citrus fruits, and obtaining the texture concavity and convexity data of different maturity areas;

[0128] Step S322: performing surface force transmission difference analysis on the texture concave-convex data of different maturity areas to obtain surface force transmission difference data;

[0129] Step S323: Calculating the surface compression force of citrus fruits at different maturity levels based on the surface force transmission difference data, and obtaining surface compression force data of citrus fruits at different maturity levels;

[0130] Step S324: performing surface force recovery and rebound analysis on the surface compression force data of different maturity areas to obtain surface force recovery and rebound data of different maturity areas;

[0131] Step S325: Perform external morphology perception analysis of different maturity areas based on the surface force recovery rebound data of different maturity areas and the texture concave-convex data of different maturity areas to obtain citrus external morphology perception data.

[0132] In an embodiment of the present invention, the texture features of different maturity areas are analyzed for concave-convex feeling according to the tactile perception data of citrus. First, the surface of citrus is scanned at multiple points using a high-precision tactile sensor to collect tactile perception data of different areas. The tactile sensor can detect tiny protrusions and depressions on the surface of citrus, and then generate a high-resolution surface height map. In order to further analyze the texture differences in different maturity areas, a mathematical morphological processing method is used to extract the tiny concave-convex features of the peel surface through convolution operations. Then, based on these concave-convex features, the Laplace operator is applied to enhance the edge and high-frequency information in the area to extract the concave-convex feeling changes on the peel surface. By analyzing the concave-convex feeling change pattern of different maturity areas, texture concave-convex feeling data is finally generated, which reflects the physical characteristic differences of the surface texture of citrus at different maturity stages. The surface force transmission difference analysis is performed on the texture concave-convex feeling data of different maturity areas. Based on the texture concave-convex feeling data, a mechanical model of the citrus surface is first established. The model uses a finite element analysis method to simulate the surface texture as the point of force application. Then, a uniform compressive force is applied to analyze the force transmission path and force distribution in different maturity areas. By observing the force transmission caused by different textures in each area and combining the elastic modulus parameters, the force changes on the surface of citrus are calculated. This process evaluates the difference in force transmission on the surface of different areas through the changes in stress-strain curves and deformation fields. Finally, surface force transmission difference data are generated, which shows the force transmission law caused by the unevenness of the surface texture in different maturity areas of citrus. According to the surface force transmission difference data, the surface compression force calculation of citrus in different maturity areas is carried out. First, based on the surface force transmission difference data obtained in the previous step, the local compression force characteristics of each maturity area are analyzed. Using the elastic mechanics theory and combining the compression modulus of the area, the compression degree of the peel in different texture areas under a certain compression force is calculated. Through the Hertz contact theory model, the relationship between the compression depth and force in different areas is further accurately described, and the compression response of the surface is quantified. In this process, the regional division is refined to accurately evaluate the local force conditions in different maturity areas. Finally, the surface compression force data are generated, which reflects the physical response of the citrus peel when it is compressed at different maturity stages. The surface force recovery and rebound analysis is performed on the surface compression force data of different maturity areas. First, the deformation recovery process of the citrus peel after the release of external force is simulated by a dynamic compression experiment, and the deformation recovery time and recovery degree of each area are recorded. Combined with the compression force data, Hooke's law and material mechanics recovery model are applied to quantitatively evaluate the rebound behavior of the skin. The specific method is: monitor the recovery speed and deformation recovery curve of the surface layer through a real-time displacement measurement instrument, and analyze the rebound characteristics of each area. Furthermore, the relationship between the recovery force and recovery time of each area after the external force is removed is calculated.Finally, the surface force recovery and rebound data are obtained, which are used to describe the elastic recovery ability of citrus peel at different maturity stages and reflect its external quality. According to the surface force recovery and rebound data and texture concave-convex data of different maturity areas, the external morphological perception analysis of different maturity areas is carried out. First, the surface force recovery and rebound data are combined with the texture concave-convex data, and the changes in surface morphology are comprehensively analyzed using the morphological filtering algorithm. This method can accurately extract the comprehensive characteristics of the epidermal morphology at different maturity stages. By calculating the correlation between the surface tension and elastic recovery of each area, the specific impact of maturity differences on the morphology of the peel is determined. Then, through edge detection and morphological operator operations, the changes in morphological contours of different areas are further analyzed, and combined with the previously generated data, the appearance differences of the citrus surface are accurately evaluated. Finally, the external morphological perception data of citrus is generated, which is used to accurately evaluate the maturity and external quality of citrus.

[0133] Preferably, step S4 comprises the following steps:

[0134] Step S41: Based on the proximal strategy optimization algorithm, the external morphology perception data of the citrus and the internal maturity grade data of the citrus are subjected to citrus internal and external quality memory learning to obtain citrus internal and external quality memory learning data;

[0135] Step S42: normalizing the citrus internal and external quality memory learning data to obtain citrus internal and external quality memory normalized data;

[0136] Step S43: Design citrus internal and external quality selection firmware based on the normalized data of citrus internal and external quality memory, obtain citrus internal and external quality selection firmware, and send the citrus internal and external quality selection firmware to the cloud platform to perform citrus internal and external quality inspection and grading.

[0137] In an embodiment of the present invention, based on the proximal policy optimization (PPO) algorithm, the external morphological perception data of citrus and the internal maturity grade data of citrus are used to perform citrus internal and external quality memory learning. First, the proximal policy optimization algorithm is used to establish a quality learning model, which optimizes the strategy through multiple iterations and updates to accurately learn the internal and external quality characteristics of citrus. The specific operation includes inputting the external morphological perception data of citrus (such as skin texture, force rebound data) and the internal maturity grade data of citrus. The proximal policy optimization algorithm is based on the reinforcement learning framework and continuously adjusts the strategy to maximize the accuracy of the matching of internal and external quality characteristics. This process enables the system to gradually learn the correlation between citrus quality characteristics through iterative updates of state-action pairs, and finally generates citrus internal and external quality memory learning data. This data reflects the system's deep understanding of the internal and external quality characteristics of citrus, laying the foundation for subsequent processing. The citrus internal and external quality memory learning data is normalized to eliminate the differences between the data caused by different scales, ensuring that subsequent analysis and processing are more stable and accurate. Normalization processing uses the minimum-maximum normalization method to compress the quality feature values ​​in the memory learning data into the range of [0,1]. This processing method avoids the influence of excessively large or small numerical values ​​of feature values, ensures that the data are compared and calculated at the same scale, and finally generates the memory normalized data of the internal and external quality of citrus, providing a unified input standard for the subsequent firmware design. Based on the memory normalized data of the internal and external quality of citrus, the firmware for selecting the internal and external quality of citrus is designed. The core of the design of this firmware is to ensure that the system can quickly classify and screen citrus based on the normalized quality data by embedding algorithm logic. First, based on the normalized data, the decision tree algorithm is used to design classification rules. The decision tree divides the internal and external quality of citrus into different levels by analyzing the hierarchical relationship of different quality features, and sets specific thresholds for the quality screening operation of the firmware. For example, when the bumpiness of the peel texture exceeds a specific value and the internal maturity reaches a certain standard, the firmware will classify the citrus as "high quality". At the same time, the hard-coded logic in the firmware will dynamically adjust the screening criteria according to the distribution of the normalized data. After the design is completed, the firmware for selecting the internal and external quality of citrus will be sent to the cloud platform, and the terminal devices on the platform will call the firmware to perform internal and external quality inspection and grading of citrus, realizing fast and accurate quality screening and grading operations.

[0138] Preferably, the present invention further provides a citrus internal and external quality detection and grading system based on deep learning, which is used to execute the citrus internal and external quality detection and grading method based on deep learning as described above, and the citrus internal and external quality detection and grading system based on deep learning comprises:

[0139] The data acquisition module is used to collect spectral images of citrus through a hyperspectral imager to obtain spectral characteristic band data; collect tactile perception data of citrus through a tactile sensor built into the mechanical claw to obtain tactile perception data of citrus;

[0140] The internal maturity grade classification module is used to identify the sugar-acid ratio gradient of different citrus regions according to the spectral characteristic band data, and obtain the citrus sugar-acid ratio gradient data; evaluate the difference in pulp elasticity of citrus in different regions according to the citrus sugar-acid ratio gradient data and the citrus tactile perception data, and obtain the citrus pulp elasticity difference data; divide the internal maturity grades of different citrus regions based on the citrus sugar-acid ratio gradient data and the citrus pulp elasticity prediction data, and obtain the citrus internal maturity grade data;

[0141] The external morphology perception module is used to perform external morphology perception analysis on the internal maturity grade data of citrus fruits in different maturity areas according to the spectral characteristic band data to obtain external morphology perception data of citrus fruits;

[0142] The firmware design module is used to learn the internal and external quality of citrus by memorizing the external morphological perception data and the internal maturity grade data of citrus based on the proximal strategy optimization algorithm, and memorize the normalized data of the internal and external quality of citrus; the firmware for selecting the internal and external quality of citrus is designed based on the normalized data of the internal and external quality of citrus, and the firmware for selecting the internal and external quality of citrus is obtained, and the firmware for selecting the internal and external quality of citrus is sent to the cloud platform to perform internal and external quality detection and grading of citrus.

[0143] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0144] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for detecting and grading the internal and external quality of citrus based on deep learning, characterized in that: The following steps are involved: Step S1: collecting spectral images of citrus fruits through a hyperspectral imager to obtain spectral characteristic band data; The tactile perception data of the citrus is collected by using the tactile sensor built into the mechanical claw to obtain the tactile perception data of the citrus; Step S2: identifying the sugar-acid ratio gradient of different regions of citrus according to the spectral characteristic band data, and obtaining the sugar-acid ratio gradient data of citrus; evaluating the difference in pulp elasticity of citrus in different regions according to the sugar-acid ratio gradient data of citrus and the tactile perception data of citrus, and obtaining the difference in pulp elasticity of citrus; dividing the internal maturity level of different regions of citrus based on the sugar-acid ratio gradient data of citrus and the citrus pulp elasticity prediction data, and obtaining the internal maturity level data of citrus; Step S3: performing external morphological perception analysis of different maturity areas on the internal maturity grade data of the citrus according to the spectral characteristic band data to obtain external morphological perception data of the citrus; Step S4: Based on the proximal strategy optimization algorithm, the external morphology perception data of the citrus and the internal maturity grade data of the citrus are used for internal and external quality memory learning of the citrus, so as to obtain normalized internal and external quality memory data of the citrus; Based on the normalized data of citrus internal and external quality memory, citrus internal and external quality selection firmware is designed to obtain citrus internal and external quality selection firmware, and the citrus internal and external quality selection firmware is sent to the cloud platform to perform citrus internal and external quality inspection and grading.

2. The method for detecting and grading the internal and external quality of citrus fruits based on deep learning according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: collecting spectral images of citrus using a hyperspectral imager to obtain spectral images of citrus; Step S12: performing image enhancement processing on the citrus spectrum image to obtain a citrus spectrum enhanced image; Step S13: extracting characteristic bands from the citrus spectrum enhanced image to obtain spectral characteristic band data; Step S14: collecting tactile perception data of the citrus through the tactile sensor built into the mechanical claw to obtain tactile perception data of the citrus.

3. The method for detecting and grading the internal and external quality of citrus fruits based on deep learning according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: using a preset spectral band similarity recognition model to classify the spectral feature band data into similar bands to obtain spectral feature band similarity data; Step S22: identifying the sugar-acid ratio gradient of different regions of citrus according to the similarity data of the spectral characteristic bands, and obtaining the sugar-acid ratio gradient data of citrus; Step S23: evaluating the difference in pulp elasticity of citrus in different areas according to the citrus sugar-acid ratio gradient data and the citrus tactile perception data, and obtaining citrus pulp elasticity difference data; Step S24: Based on the citrus sugar-acid ratio gradient data and the citrus pulp elasticity difference data, the internal maturity grade of different citrus areas is divided to obtain the citrus internal maturity grade data.

4. The method for detecting and grading the internal and external quality of citrus fruits based on deep learning according to claim 3, characterized in that: Step S22 includes the following steps: Step S221: extracting sugar and acid bands of different regions of citrus from similar data of spectral characteristic bands to obtain sugar and acid spectral band data; Step S222: performing multi-region sugar-acid distribution chromatography processing on different regions of citrus based on the sugar-acid spectral band data to obtain multi-region sugar-acid distribution chromatography data; Step S223: performing band density clustering calculation on the sugar-acid spectrum band data of different regions of citrus according to the multi-region sugar-acid distribution chromatography data to obtain sugar-acid distribution band density clustering data; Step S224: identifying the sugar-acid ratio gradients of different citrus regions using the sugar-acid distribution band density clustering data to obtain citrus sugar-acid ratio gradient data.

5. The method for detecting and grading the internal and external quality of citrus fruits based on deep learning according to claim 3, characterized in that: Step S23 includes the following steps: Step S231: calculating the elastic modulus of different regions of the citrus based on the tactile perception data of the citrus to obtain regional elastic modulus data of the citrus; Step S232: extracting the surface temperature and humidity of different areas of the citrus from the tactile perception data of the citrus to obtain the surface temperature and humidity data of the citrus area; Step S233: Calculate the elastic tension of different citrus regions according to the elastic modulus data of the citrus regions and the surface temperature and humidity data of the citrus regions to obtain the elastic tension data of the citrus regions; Step S234: performing periodic deposition prediction of different citrus regions on the citrus sugar-acid ratio gradient data to obtain citrus sugar-acid ratio periodic deposition data; Step S235: Evaluate the difference in pulp elasticity of different regions of citrus according to the citrus sugar-acid ratio periodic deposition data and the citrus regional elastic tension data to obtain citrus pulp elasticity difference data.

6. The method for detecting and grading the internal and external quality of citrus fruits based on deep learning according to claim 4, characterized in that: The periodic sedimentation prediction of different citrus regions based on the citrus sugar-acid ratio gradient data includes the following steps: Based on the citrus sugar-acid ratio gradient data, the citrus cell osmotic pressure change data of different citrus regions were analyzed to obtain the citrus cell osmotic pressure change data; According to the osmotic pressure change data of citrus cells, the cell wall expansion force simulation calculation of different citrus regions was performed to obtain the cell wall expansion force simulation data; The cell membrane permeability data was obtained by analyzing the cell membrane permeability according to the osmotic pressure change data of citrus cells and the simulated data of cell wall swelling force; Based on the citrus cell osmotic pressure change data and cell membrane permeability data, the periodic deposition of different citrus regions was predicted, and the periodic deposition data of the citrus sugar-acid ratio were obtained.

7. The method for detecting and grading the internal and external quality of citrus fruits based on deep learning according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: analyzing the texture characteristics of citrus in different maturity areas on the internal maturity grade data of citrus according to the spectral characteristic band data to obtain maturity difference texture characteristic data; Step S32: performing external morphological perception analysis of different maturity areas on the maturity difference texture feature data according to the tactile perception data of the citrus to obtain external morphological perception data of the citrus.

8. The method for detecting and grading the internal and external quality of citrus fruits based on deep learning according to claim 7, characterized in that: Step S32 includes the following steps: Step S321: analyzing the texture concavity and convexity of different maturity areas on the maturity difference texture feature data according to the tactile perception data of the citrus fruits, and obtaining the texture concavity and convexity data of different maturity areas; Step S322: performing surface force transmission difference analysis on the texture concave-convex data of different maturity areas to obtain surface force transmission difference data; Step S323: Calculating the surface compression force of citrus fruits at different maturity levels based on the surface force transmission difference data, and obtaining surface compression force data of citrus fruits at different maturity levels; Step S324: performing surface force recovery and rebound analysis on the surface compression force data of different maturity areas to obtain surface force recovery and rebound data of different maturity areas; Step S325: Perform external morphology perception analysis of different maturity areas based on the surface force recovery rebound data of different maturity areas and the texture concave-convex data of different maturity areas to obtain citrus external morphology perception data.

9. The method for detecting and grading the internal and external quality of citrus fruits based on deep learning according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: Based on the proximal strategy optimization algorithm, the external morphology perception data of the citrus and the internal maturity grade data of the citrus are subjected to citrus internal and external quality memory learning to obtain citrus internal and external quality memory learning data; Step S42: normalizing the citrus internal and external quality memory learning data to obtain citrus internal and external quality memory normalized data; Step S43: Design citrus internal and external quality selection firmware based on the normalized data of citrus internal and external quality memory, obtain citrus internal and external quality selection firmware, and send the citrus internal and external quality selection firmware to the cloud platform to perform citrus internal and external quality inspection and grading.

10. A citrus internal and external quality detection and grading system based on deep learning, characterized in that: Used to execute the citrus internal and external quality detection and grading method based on deep learning as claimed in claim 1, the citrus internal and external quality detection and grading system based on deep learning comprises: The data acquisition module is used to collect spectral images of citrus through a hyperspectral imager to obtain spectral characteristic band data; collect tactile perception data of citrus through a tactile sensor built into the mechanical claw to obtain tactile perception data of citrus; The internal maturity grade classification module is used to identify the sugar-acid ratio gradient of different citrus regions according to the spectral characteristic band data, and obtain the citrus sugar-acid ratio gradient data; evaluate the difference in pulp elasticity of citrus in different regions according to the citrus sugar-acid ratio gradient data and the citrus tactile perception data, and obtain the citrus pulp elasticity difference data; divide the internal maturity grades of different citrus regions based on the citrus sugar-acid ratio gradient data and the citrus pulp elasticity prediction data, and obtain the citrus internal maturity grade data; The external morphology perception module is used to perform external morphology perception analysis on the internal maturity grade data of citrus fruits in different maturity areas according to the spectral characteristic band data to obtain external morphology perception data of citrus fruits; The firmware design module is used to learn the internal and external quality of citrus by memorizing the external morphological perception data and the internal maturity grade data of citrus based on the proximal strategy optimization algorithm, and memorize the normalized data of the internal and external quality of citrus; the firmware for selecting the internal and external quality of citrus is designed based on the normalized data of the internal and external quality of citrus, and the firmware for selecting the internal and external quality of citrus is obtained, and the firmware for selecting the internal and external quality of citrus is sent to the cloud platform to perform internal and external quality detection and grading of citrus.

Images