A three-dimensional machine vision-based fragrance mixing dual-arm robot system

Abstract

The present application relates to the technical field of aromatherapy preparation, and proposes a fragrance preparation double-arm robot system based on three-dimensional machine vision, which comprises a fragrance preparation double-arm robot, the fragrance preparation double-arm robot is provided with a three-dimensional machine vision module, a fragrance preparation algorithm module and a mechanical arm provided with a clamping piece; the fragrance preparation algorithm module is used for obtaining the aromatherapy demand of a user, generating a plurality of aromatherapy recipes based on the characteristics of aromatherapy raw materials and the aromatherapy demand of the user for the user to select; the three-dimensional machine vision module is used for identifying and positioning the positions of the required aromatherapy raw materials and the fragrance preparation container after the user selects at least one aromatherapy recipe, and generating fragrance preparation instructions for driving the mechanical arm and the clamping piece; the fragrance preparation instructions are used for driving the mechanical arm and the clamping piece to perform weighing, feeding, liquid transfer and stirring operations according to the positions of the required aromatherapy raw materials and the fragrance preparation container, the setting proportion and the preparation sequence of the aromatherapy recipe, so as to obtain prepared aromatherapy, bottle and package the prepared aromatherapy and output the prepared aromatherapy to the user.

Description

Technical Field

[0001] The present invention relates to the technical field of aromatherapy modulation, in particular to a fragrance-adjusting dual-arm robot system based on three-dimensional machine vision. Background Art

[0002] As consumers pursue personalized, high-quality lifestyles, the fragrance industry needs to continuously launch innovative products that meet market demands. Traditional aromatherapy creation typically relies on the perfumer's experience and manual labor. The fragrance-making process involves selecting, proportioning, and mixing raw materials. This process often requires multiple trials and adjustments to ensure the perfect concentration and aroma. Due to its high dependency and labor intensity, the process is inefficient. Perfumers constantly evaluate the aroma and make adjustments as needed. This process requires repeated experimentation to achieve the desired effect.

[0003] Therefore, traditional aromatherapy mainly relies on the experience of perfumers, and consumers are usually unable to participate in the specific aromatherapy process. They can only choose ready-made fragrances or rely on the suggestions of perfumers.

[0004] Some businesses now allow consumers to customize fragrances on site, but since individual consumers lack experience in perfume making, it is difficult for them to make a satisfactory fragrance on their own. Summary of the Invention

[0005] In response to the above-mentioned defects, the purpose of the present invention is to propose a fragrance-adjusting dual-arm robot system based on three-dimensional machine vision, which aims to intelligently generate multiple aromatherapy formulas that meet user needs, provide highly personalized aromatherapy modulation services, and meet the personalized needs of different users.

[0006] To achieve this object, the present invention adopts the following technical solutions:

[0007] A fragrance-adjusting dual-arm robot system based on three-dimensional machine vision, comprising the fragrance-adjusting dual-arm robot, wherein the fragrance-adjusting dual-arm robot is configured with a three-dimensional machine vision module, a fragrance-adjusting algorithm module, and a robotic arm equipped with a clamping member;

[0008] The aromatherapy algorithm module is used to obtain the user's aromatherapy needs and generate several aromatherapy formulas for the user to choose based on the characteristics of the aromatherapy raw materials and the user's aromatherapy needs;

[0009] The three-dimensional machine vision module is used to identify and locate the required aromatherapy raw materials and fragrance containers after the user selects at least one aromatherapy formula, and generate fragrance instructions to drive the robotic arm and the gripper;

[0010] The fragrance blending instruction is used to drive the mechanical arm and the gripper to perform the weighing, feeding, liquid transferring and stirring operations according to the required fragrance raw materials, the positions of the fragrance blending containers, the fragrance formula setting proportions and the fragrance blending sequence, so as to obtain the blended fragrance, pack the blended fragrance into bottles and output the blended fragrance to the user.

[0011] Preferably, the fragrance blending algorithm module is used to receive the fragrance demand input by the user, generate a plurality of fragrance formulas and visually output the fragrance formulas to the user for confirmation.

[0012] The fragrance demand includes the fragrance tone level, the concentration, the odor type, the use occasion, the season and the duration.

[0013] Preferably, the three-dimensional machine vision module is used to pre-paste labels on the fragrance raw materials and the fragrance blending containers, and after the user selects at least one fragrance formula, the corresponding labels are searched according to the fragrance formula selected by the user to determine the positions of the required fragrance raw materials and the fragrance blending containers, and the fragrance blending instruction driving the mechanical arm and the gripper is generated.

[0014] After the user feeds back the formula, incremental learning is performed based on the evaluation of the user to optimize the generation of the next fragrance formula.

[0015] Further, after the mechanical arm and the gripper receive the fragrance blending instruction, the position of the mechanical arm moving with the gripper changes according to the fragrance blending sequence, so that the gripper moves to the position of the required fragrance raw material, the gripper clamps the required fragrance raw material according to the fragrance formula setting proportion, the clamped fragrance raw material is weighed and then added to the fragrance blending container, until all the required fragrance raw materials are added to the fragrance blending container according to the fragrance formula setting proportion.

[0016] The gripper clamps the stirring tool to stir the fragrance raw materials in the fragrance blending container to obtain the blended fragrance, and then the blended fragrance is added to different fragrance bottles according to the net content of the fragrance selected by the user, and the user is reminded to pick up the fragrance.

[0017] Preferably, in the fragrance blending algorithm module, generating a plurality of fragrance formulas based on the characteristics of the fragrance raw materials and the fragrance demand of the user includes:

[0018] Converting the fragrance demand of the user into a feature vector;

[0019] Using clustering analysis and decision tree to establish a user preference model, and determining the user type based on the feature vector;

[0020] Generating preselected fragrance formulas according to the characteristics of the raw materials and the feature vector, and generating a plurality of alternative fragrance formulas according to the user type using a collaborative filtering scheme;

[0021] Iteratively optimizing the preselected fragrance formulas and the alternative fragrance formulas to obtain a plurality of preferred fragrance formulas.

[0022] Further, in the fragrance blending algorithm module, a user preference model is established using cluster analysis and decision tree, and a user type is determined based on a feature vector, including:

[0023] Users with similar preferences are classified into different types using cluster analysis;

[0024] The feature mean of each user type is analyzed to obtain the common preferences of each group;

[0025] The feature vector is input into the decision tree as the user feature, and the importance of the user feature is analyzed;

[0026] The fragrance formula trend is obtained by combining the user type and the results of the importance analysis.

[0027] Further, in the fragrance blending algorithm module, preselected fragrance formulas are generated based on the characteristics of the raw materials and the feature vector, including:

[0028] The characteristics of all raw materials are obtained;

[0029] The combination of raw materials is obtained based on the characteristics of the raw materials and the fragrance note levels required by the user;

[0030] The conventional blending ratio is adjusted according to the user's concentration requirement to obtain the latest concentration ratio;

[0031] The combination of raw materials and the latest concentration ratio are combined to obtain the preselected fragrance formula.

[0032] Further, in the fragrance blending algorithm module, a number of alternative fragrance formulas are generated according to the user type using a collaborative filtering scheme, including:

[0033] The fragrance formula is recommended to the user by combining the formula selection of other users of the same user type and the fragrance formula trend;

[0034] A number of first alternative fragrance formulas are obtained by combining the user type and using a collaborative filtering algorithm on the recommended fragrance formula;

[0035] The first alternative fragrance formulas are filtered and selected based on the user's requirements to obtain a number of second alternative fragrance formulas.

[0036] Preferably, the fragrance blending instructions include a fragrance combination code, which includes the required fragrance raw materials and the fragrance formula setting ratio.

[0037] One of the above technical solutions has the following advantages or beneficial effects:

[0038] The application can generate a fragrance formula according to the personalized needs of a user based on three-dimensional machine vision, realizes high-precision raw material identification and blending through real-time identification of the position of raw materials by a three-dimensional machine vision module and the collaborative operation of the three-dimensional machine vision module and a mechanical arm, has high automation, ensures accurate grabbing and dispensing of the mechanical arm, avoids errors in manual blending, improves the accuracy of blending, and improves production efficiency and product quality; the blending algorithm module provides a plurality of preferred formulas according to the characteristics of the spices and user preferences, ensures high consistency and high quality of the fragrance products; can monitor the raw material inventory and usage in real time, reduce raw material waste caused by misoperation, reduce production costs, support users to select different fragrance contents and bottle styles, flexibly meet the customization needs of users, while ensuring the accuracy and quality of sample blending, in addition, can be extended to other fields that require accurate proportioning, such as cosmetics, soap, etc., to realize wide cross-field application. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, a brief introduction will be given below to the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of the provided drawings.

[0040] Figure 1 is the actual application diagram of the fragrance blending dual-arm robot based on three-dimensional machine vision provided by the embodiment of the present application;

[0041] Figure 2 is the fragrance blending flowchart of the fragrance blending dual-arm robot based on three-dimensional machine vision provided by the embodiment of the present application;

[0042] Among them, the three-dimensional machine vision module 1, the mechanical arm 2, the clamping piece 3. DETAILED DESCRIPTION

[0043] The embodiments of the present application will be described in detail below, and the examples of the embodiments are shown in the drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are only used to explain the present application, and cannot be understood as a limitation of the present application.

[0044] In the present invention, the terms "comprising", "containing" or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements not only includes those elements, but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without more limitations, the element defined by the phrase "comprising a" does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0045] With the pursuit of individualization and high-quality life by consumers, the fragrance industry needs to continuously launch innovative products that meet market demand. Traditional fragrance modulation usually relies on the experience and manual operation of fragrance modulators, and the fragrance modulation process includes the selection, proportioning and mixing of raw materials. The fragrance modulation process often needs to be tested and adjusted several times to ensure the concentration and fragrance of the incense. Due to its high dependence and labor intensity, the efficiency is low. The fragrance modulator will continuously evaluate the fragrance of the incense and make adjustments as needed. This process requires repeated testing to achieve the desired effect.

[0046] As shown in Figure 1 A three-dimensional machine vision-based fragrance modulation dual-arm robot system, comprising the fragrance modulation dual-arm robot, the fragrance modulation dual-arm robot is configured with a three-dimensional machine vision module 1, a fragrance modulation algorithm module and a mechanical arm 2 provided with a clamping piece 3;

[0047] The fragrance modulation algorithm module is used to obtain the aromatherapy needs of the user, and generate a plurality of aromatherapy formulas based on the characteristics of the aromatherapy raw materials and the aromatherapy needs of the user for the user to choose;

[0048] The three-dimensional machine vision module 1 is used to identify and locate the positions of the required aromatherapy raw materials and the fragrance modulation container after the user selects at least one aromatherapy formula, and generate fragrance modulation instructions for driving the mechanical arm 2 and the clamping piece 3;

[0049] The fragrance modulation instructions are used to drive the mechanical arm 2 and the clamping piece 3 to perform weighing, feeding, liquid transfer and stirring operations according to the positions of the required aromatherapy raw materials and the fragrance modulation container, the setting proportion and the fragrance modulation sequence of the aromatherapy formula, so as to obtain the modulated aromatherapy. The modulated aromatherapy is bottled, packaged and output to the user.

[0050] The aromatherapy needs of the user refer to the preferences of the user when selecting aromatherapy, including but not limited to fragrance type, concentration, duration, application occasion (such as daily use, relaxation, sleep, etc.), and seasonal preference (such as fresh fragrance type in summer and warm fragrance type in winter) and the like.

[0051] The raw materials of aromatherapy are diverse, each raw material has its unique aroma characteristics, volatility, concentration, duration and other properties, the extraction of raw material characteristics is the basis for formulating aromatherapy formula, if the user selects the "front note" preference as citrus flavor, a combination containing orange, lemon and other spices will be recommended, referring to the characteristic vector, that is, referring to the user's needs, a preliminary aromatherapy formula can be obtained, which includes the raw materials and corresponding concentrations of the formula.

[0052] Each robotic arm 2 of the fragrance blending robot can perform different tasks. For example, one robotic arm 2 is used to grab raw materials and weigh them, and the other robotic arm 2 can be responsible for stirring or transferring liquids; for example, one robotic arm 2 can be used to grab fragrance containers such as fragrance bottles, and the other robotic arm 2 can grab extraction tubes to extract raw material fragrance liquid into the fragrance bottle. The movement of the robotic arm 2 can be driven by a stepper motor or a servo motor, combined with precise motion control algorithms to ensure the accuracy and coordination of each operation. The three-dimensional machine vision module 1 is the "eyes" of the system, which captures images of the surrounding environment through cameras, lidar, infrared sensors and other devices, and analyzes the three-dimensional spatial position and shape of objects according to computer vision algorithms. The vision module can identify and locate fragrance raw materials and containers through deep learning algorithms, such as deep learning models (e.g. convolutional neural networks CNN) that can classify and locate objects based on their features (e.g. color, shape, texture, etc.). The machine vision system can calculate the coordinates of objects in three-dimensional space (XYZ coordinates) to provide accurate position data for subsequent robotic arm 2 operations.

[0053] Subsequently, the fragrance blending algorithm module analyzes the user's needs and generates multiple preferred fragrance blending formulas in combination with the characteristics of the aromatherapy raw materials. For example, a raw material database can be pre-established, which should include the name, characteristics, odor description, use suggestion, interaction, etc. of each spice, and finally generate several preferred formulas that meet the user's needs, each formula includes raw materials and their use ratio, finally for the user to choose.

[0054] The three-dimensional machine vision module 1 is responsible for accurately identifying the positions of the aromatherapy raw materials and the blending containers, and provides real-time visual feedback for the robot. In this embodiment, the aromatherapy blending robot can be equipped with a blending area, which is configured with positions for placing the blended aromatherapy products and various spices. After receiving the user-selected aromatherapy formula, the three-dimensional machine vision module 1 identifies and locates the positions of the required aromatherapy raw materials and blending containers, and generates blending instructions for driving the mechanical arm 2 and the gripper 3. When the mechanical arm 2 and the gripper 3 receive the blending instructions, the required weight of each aromatherapy raw material is calculated according to the blending ratio set in the blending instructions. The gripper 3 picks up the aromatherapy raw material for weighing. After weighing the required weight, the mechanical arm 2 extracts the specified spice through the gripper 3 and transfers it to the blending container according to the feeding ratio. After all the spices are added to the blending container, the left arm of the aromatherapy blending robot can grab a pipette, and the right arm can grab a test tube containing spices. According to the user's preferred aromatherapy formula, the required proportion of spices is sequentially added to the test tube in order. The blended aromatherapy can be divided into several aromatherapy products with different contents, for example, according to the user's selection of the incense net content, the blended incense is automatically injected into the specified incense bottle, and after the bottling is completed, the user is prompted to take the incense.

[0055] Based on three-dimensional machine vision, the present application can generate aromatherapy formulas according to the user's personalized needs, real-time identify the raw material position through the three-dimensional machine vision module 1, and realize high-precision raw material identification and blending through the collaborative operation of the three-dimensional machine vision module 1 and the mechanical arm 2, which has high automation, ensures the accurate grabbing and dispensing of the mechanical arm 2, avoids the errors of manual blending, improves the accuracy of blending, and improves the production efficiency and product quality. The formula generation module provides multiple preferred formulas according to the characteristics of the spices and the user's preferences, ensuring the high consistency and high quality of the aromatherapy products. It can monitor the raw material inventory and usage in real time, reduce the waste of raw materials caused by misoperation, reduce the production cost, support the user to select different incense net contents and bottle styles, flexibly meet the user's customized needs, while ensuring the accuracy and quality of sample blending. In addition, as shown in the aromatherapy blending robot, it is not only suitable for incense blending, but also can be extended to other fields that require accurate proportioning, such as cosmetics, soap, etc., realizing wide cross-field application. Figure 1

[0056] Preferably, the blending algorithm module is used to receive the user's input aromatherapy needs, generate several aromatherapy formulas, and visually output the aromatherapy formulas to the user for confirmation.

[0057] The aromatherapy needs include the aroma level, concentration, odor type, use occasion, season, and duration.

[0058] ​The fragrance is the main component of the incense, usually divided into three layers: the front note (the initial aroma), the middle note (the core of the aroma) and the base note (the lasting aftertaste of the aroma), users can specify what combination of fragrance levels they want the incense to have, such as floral, fruity, woody, herbal, etc. The concentration refers to the strength or intensity of the incense, users can choose different concentrations of incense such as "fresh" and "intense", which will affect the proportion of different spices used in the incense formula. Users can choose the classification of the smell, such as floral, fruity, woody, herbal, citrus, oriental, etc., different smell types represent different combinations and arrangements of spices. The use of incense determines the selection and collocation of spices in the formula, for example, incense suitable for home may tend to warm and soothing aroma, while incense for office or public places may choose fresh and refreshing spices. The seasonality requirement of incense is also an important factor in the process of fragrance blending, winter incense tends to be warm and stable (such as woody, spice), while summer tends to be fresh and light (such as citrus, floral). The duration of the incense refers to the time the aroma remains in the air, users can choose "long-lasting" or "fast-evaporating", which will affect the proportion of volatile and long-lasting ingredients in the incense formula.

[0059] In this embodiment, the fragrance blending algorithm module analyzes the user's input incense requirements and maps these requirements to the available spice database, each spice has different smell, volatility, concentration, duration, etc. The algorithm will match according to these attributes, for example, floral spices are suitable for front notes, while woody spices are usually used for base notes. According to the different needs of the user, the algorithm will assign a certain weight to each requirement (fragrance level, concentration, smell type, season, etc.), if the user emphasizes "duration" especially, the algorithm will increase the ingredients with long-lasting aroma time (such as woody, herbal spices) in the incense formula. After generating several incense formulas, the fragrance blending robot system visualizes the incense formula to the user through a graphical interface or interactive interface, such as showing the position and proportion of different spices in the front, middle and base notes, users can see the name, concentration and volatility characteristics of each layer of spices, through labels or color coding, intuitively show the user the classification of the aroma, such as floral, fruity, woody, etc., show the distribution of the concentration and duration of the incense, help users intuitively understand the aroma intensity and duration of each formula. Users can adjust some parameters of the formula through the interface (such as increasing or decreasing the proportion of a certain spice), and real-time view the changes of the visualization results, finally confirm the most suitable incense formula. Once the user selects a final formula, the fragrance blending algorithm module will output this formula as specific formula data, generate fragrance blending instructions for subsequent operations.

[0060] Preferably, the three-dimensional machine vision module 1 is used to label the aromatherapy raw materials and the fragrance containers in advance, and after the user selects at least one aromatherapy formula, the corresponding label is searched according to the aromatherapy formula selected by the user to determine the positions of the required aromatherapy raw materials and the fragrance containers, and the fragrance mixing instructions for driving the mechanical arm 2 and the gripper 3 are generated; after the user feeds back the formula, incremental learning is performed based on the user's evaluation to optimize the generation of the next aromatherapy formula.

[0061] Each of the aromatherapy raw materials and containers is labeled with a label (such as a two-dimensional code, a bar code, or a visual label) having a unique identification, and after the user selects an aromatherapy formula, the system identifies the required label according to the aromatherapy raw material and container information in the formula, the three-dimensional machine vision module 1 identifies the label of the aromatherapy raw material and the container, and locates the specific position of the aromatherapy raw material and the container in the workspace. Through the identification of the label, the system can accurately know the current positions of the aromatherapy raw materials and the containers, and then transmit these position information to the system, determine the actions required to be performed by the mechanical arm 2 according to the identified label information, for example, which bottle of aromatherapy raw material or which container needs to be grabbed by the gripper, calculate the optimal path for the mechanical arm 2 to reach the target raw material or container, and consider the obstacles in the space, the arrangement of the raw materials, and other factors.

[0062] Further, after the mechanical arm 2 and the gripper 3 receive the fragrance mixing instructions, the position of the mechanical arm 2 is changed according to the fragrance mixing sequence to move the gripper 3 to the position of the required aromatherapy raw material, the gripper 3 takes the required aromatherapy raw material according to the set proportion of the aromatherapy formula, weighs the taken aromatherapy raw material, and then adds the weighed aromatherapy raw material into the fragrance mixing container until all the required aromatherapy raw materials are added into the fragrance mixing container according to the set proportion of the aromatherapy formula.

[0063] The gripper 3 takes the stirring tool to stir the aromatherapy raw materials in the fragrance mixing container to obtain the prepared aromatherapy, and then adds the prepared aromatherapy into different aromatherapy bottles according to the net content of the aromatherapy selected by the user, and reminds the user to take it.

[0064] When the mechanical arm 2 and the clamping piece 3 receive the fragrance instruction, the mechanical arm 2 is controlled to move and drive the clamping piece 3 to adjust to the position of the required aromatherapy raw material according to the sequence set by the fragrance formula. The clamping piece 3 accurately clamps the required aromatherapy raw material according to the set proportion in the formula, and weighs it to ensure the accurate amount of each raw material. After the weighing is completed, the clamping piece 3 adds the aromatherapy raw material to the fragrance container one by one until all the raw materials are added according to the proportion of the aromatherapy formula. Then, the clamping piece 3 takes out the stirring tool and stirs the aromatherapy raw materials in the container until the required aromatherapy is prepared. Finally, the system distributes the prepared aromatherapy evenly into different aromatherapy bottles according to the net content of the aromatherapy selected by the user, and reminds the user to take the prepared aromatherapy product after completion. The whole process is efficient and accurate, and ensures the quality of each bottle of aromatherapy and the personalized needs of the user.

[0065] In the fragrance algorithm module, a plurality of fragrance formulas are generated based on the characteristics of the aromatherapy raw materials and the user's aromatherapy needs, including:

[0066] The user's aromatherapy needs are converted into a feature vector;

[0067] A user preference model is established using cluster analysis and decision trees, and the user type is determined based on the feature vector;

[0068] Preselected fragrance formulas are generated according to the characteristics of the raw materials and the feature vector, and a plurality of alternative fragrance formulas are generated according to the user type using a collaborative filtering scheme;

[0069] The preselected fragrance formulas and the alternative fragrance formulas are iteratively optimized to obtain a plurality of preferred fragrance formulas.

[0070] The user's inputted aromatherapy requirements are often unstructured, which is difficult to handle directly in the traditional way, so it needs to be converted into a form that the machine can understand through some methods, so it is converted into a feature vector, through which we can convert the user's requirement information into a digital format for further analysis and calculation, each feature corresponds to a different dimension of the user's preference, for example: fragrance level: top note: citrus = 1, middle note: floral = 2, base note: woody = 3; concentration: low = 1, medium = 2, high = 3; odor type: fresh and natural = 1 (simplified, map this description to a numerical value); usage occasion: home = 1, office = 2, outdoor = 3; season: spring = 1, summer = 2, autumn = 3, winter = 4; duration: short = 1, medium = 2, long = 3. Assuming the user input is: top note: citrus (such as lemon), middle note: floral (such as rose), base note: woody (such as sandalwood); concentration: medium; odor type: hope it is a fresh and natural combination; usage occasion: home, especially living room; season: spring; duration: hope the fragrance can last for a long time, the feature vector is: [1, 2, 3, 2, 1, 1, 1, 3]. Clustering analysis is a method of unsupervised learning, aiming to group data according to their similarity. Clustering analysis calculates the similarity between data points (such as using Euclidean distance, cosine similarity, etc.), and groups users with similar requirements into the same category. Clustering analysis determines the similarity between data points by selecting appropriate metrics, such as calculating the Euclidean distance between user requirement feature vectors. If the distance is small, it means that the two users' requirements are similar and can be grouped into the same category. Common algorithms used in clustering analysis include K-means and DBSCAN, which automatically discover the inherent structure in user requirements and then divide users into different groups. Through clustering analysis, we can identify user groups with similar requirements and recommend similar aromatherapy formulas to these users. Decision trees are a classification and regression method in supervised learning, which classifies users based on features. Its working principle is to split based on a series of conditions (features) until it reaches a leaf node, which represents the predicted result. Decision trees split nodes by selecting features that maximize "information gain" or "Gini coefficient", gradually assigning users to different categories. Information gain measures the impact of a feature on the classification result, with a larger information gain indicating that the feature has a greater effect on dividing data. In aromatherapy formulation, a decision tree model can predict the type of aromatherapy that a user is most likely to like based on their features (such as fragrance preference, concentration preference, etc.). Through clustering analysis and decision trees, we can assign each user a "type" that represents their preference pattern in aromatherapy selection, such as "floral concentration preference" or "fresh fragrance preference". The determination of user types helps the subsequent recommendation system to provide customized aromatherapy formulas for users.Collaborative filtering is a recommendation algorithm based on user historical behavior or similar user group behavior, its basic principle is: if the preferences of user A and user B are similar, then the fragrance formula they like is also likely to be similar, so the previous step determines the type of the user, in this step, the formula selected by the same type of user needs to be analyzed, if user A and user B have similar choices, and user B selects a specific formula, it is considered to recommend to user A, therefore, in the process of fragrance modulation, collaborative filtering recommends multiple possible fragrance formulas to each user according to the type of different users (i.e. demand). Iterative optimization of preselected fragrance formulas and alternative fragrance formulas can be processed by genetic algorithm, which includes selecting initial population: selecting preselected fragrance formulas and alternative fragrance formulas as initial population; fitness evaluation: calculating the fitness of each formula in the initial population; crossover and mutation: selecting high fitness formulas as parents and generating first newborn formula by crossover, randomly mutating the first newborn formula to get the second newborn formula; repeating fitness evaluation and crossover mutation until reaching the preset iteration number to get several preferred fragrance formulas and their usage instructions. First, select a certain number of formulas from preselected fragrance formulas and alternative fragrance formulas as initial population. Each fragrance formula can be regarded as a chromosome, containing information such as fragrance type, proportion and odor type. For example, a certain fragrance formula may contain lavender, sandalwood and citrus, and the proportion of spices is 50%:30%:20%. Then evaluate the fitness of each formula, which can reflect the quality of the formula, and the fitness function can be based on the following indicators: fragrance level, durability, use occasion and concentration, etc. After fitness evaluation, select high fitness fragrance formulas as parents to generate the next generation of formulas, common selection methods include roulette selection, tournament selection, etc. The higher the fitness value, the higher the probability of being selected. For example, if the fitness of a formula is 90%, it has a high probability of being selected as a parent. Select two high fitness fragrance formulas for crossover to simulate the process of gene recombination, the crossover method can be to exchange part of the components of the parent to generate a new formula.For example, the cross of formula A and formula B: new formula: top note 30%, middle note 50%, base note 20%, the cross operation can be performed in multiple aspects such as perfume type, proportion, aroma characteristics, etc., the cross operation will increase the diversity of the formula, and a certain random variation is performed on the basis of the generated new formula, the variation can be realized by changing the proportion of the perfume, adding or removing a certain perfume, for example, the second new formula of the new formula after variation is top note 31%, middle note 49%, base note 20%, then the iteration of the second new formula is performed, the fitness evaluation and cross variation operation are repeatedly performed, multiple iterations are performed, new formulas are generated in each generation, and the fitness, selection, cross and variation are continuously evaluated, the number of iterations is usually preset, for example, 50 generations or 100 generations, or until a preset threshold is reached, after multiple iterations, the genetic algorithm outputs several preferred aromatherapy formulas, each preferred aromatherapy formula is attached with corresponding use instructions, the instructions include specific ingredients of the formula, suitable use occasions, aroma intensity, durability, etc., to help users better select and use aromatherapy.

[0071] wherein the calculation of fitness satisfies the relationship:

[0072] Fitness=w1*P top *S top +w2*P middle *S middle +w3*P base *S base +w4*B score +w5*D score ;

[0073] wherein w1 represents the weight coefficient of the top note, w2 represents the weight coefficient of the middle note, w3 represents the weight coefficient of the base note, w4 represents the weight coefficient of the balance, and w5 represents the weight coefficient of the duration, P top represents the proportion of the top note, P middle represents the proportion of the middle note, P base represents the proportion of the base note, S top represents the score of the top note, S middle represents the score of the middle note, S base represents the score of the base note, B score represents the balance score, and D score represents the duration score.

[0074] Further, in the perfume blending algorithm module, a user preference model is established using cluster analysis and decision tree, and a user type is determined based on a feature vector, comprising:

[0075] using cluster analysis to divide users with similar preferences into different types;

[0076] analyzing the feature mean of each user type to obtain the common preference of each group;

[0077] input the feature vector as user features into the decision tree, and perform importance analysis on the user features;

[0078] Combine the user type and the result of the importance analysis to obtain the fragrance formula trend.

[0079] Specifically, in another embodiment, the clustering analysis can use the K-means algorithm, which is a clustering algorithm for unsupervised learning, and divide users of the same type into the same cluster. By minimizing the difference between samples in the cluster, the data in each cluster is as similar as possible, while the data difference between different clusters is as large as possible, so as to understand the common fragrance preferences of different user groups. For example, in K-means clustering, the "features" of fragrance preferences refer to fragrance notes (floral, woody, fruity, etc.), concentration (light, strong, etc.), and use occasions (daytime, nighttime, special occasions, etc.). The fragrance preferences of each user can be represented by these features. The similarity between data points is measured using the Euclidean distance. In fragrance preference analysis, the closer the distance between users, the more similar their fragrance preferences. According to this principle, the demand of each user can be calculated to distinguish user types, such as type A, which tends to prefer fresh fragrance notes or warm woody fragrance.

[0080] input the feature vector as user features into the decision tree, and perform importance analysis on the user features;

[0081] Further, in the fragrance blending algorithm module, generating preselected fragrance formulas according to the characteristics of raw materials and the feature vector comprises:

[0082] obtaining the characteristics of all raw materials;

[0083] obtaining the combination of raw materials based on the characteristics of the raw materials and the fragrance note level required by the user;

[0084] adjusting the conventional blending ratio according to the concentration requirement of the user to obtain the latest concentration ratio;

[0085] Combine the raw material combination and the latest concentration ratio to obtain a pre-selected aromatherapy formula.

[0086] Before performing the present modulation method, a database can be constructed, which contains information of each raw material, such as fragrance, odor characteristics, suitable scenes, and seasonal use recommendations, etc. The characteristics of the raw materials can be obtained from the database when needed, which facilitates subsequent aromatherapy modulation.

[0087] According to the basic theory of perfume blending, the fragrance of perfume is divided into three levels: front note, middle note and back note. Each level of fragrance has different volatile characteristics and odor intensity. Therefore, suitable fragrance combinations need to be selected from the fragrance database according to the user's fragrance level preference. The volatile speed of front note fragrance is relatively fast, and it is usually the first perceived odor. Front note usually chooses strong volatile and fresh odor, such as citrus (orange, lemon) or fresh mint, herbal plants. Middle note, also known as heart note, appears a few minutes to a few hours after the perfume is sprayed, and is usually softer and has a certain level of feeling. Common fragrances include floral (such as rose, jasmine) and vanilla, herbs. Back note is the base tone of the perfume, and the odor is the most persistent. It is usually a relatively warm and deep fragrance, such as woody fragrance (sandalwood, cedar) and spices (such as vanilla, amber), etc. For example, if the user selects "front note" preference as citrus, a combination containing orange, lemon and other fragrances will be recommended.

[0088] The concentration ratio in the aromatherapy formula determines the balance of the fragrance in each fragrance level. The common blending ratio is: 30% front note, 50% middle note and 20% back note. However, different raw material types and user needs may require different concentration ratios. After determining the combination of raw materials in the previous step, the most suitable raw material ratio is given according to user needs to ensure that the blended aromatherapy reaches the best balance in terms of fragrance, and the combination of raw materials and concentration ratio is combined to obtain a pre-selected aromatherapy formula.

[0089] Further, in the fragrance blending algorithm module, a plurality of candidate aromatherapy formulas are generated according to user types using a collaborative filtering scheme, comprising:

[0090] Combine the formula selection and aromatherapy formula trend of other users of the same user type to recommend an aromatherapy formula to the user;

[0091] Combine the user type and use a collaborative filtering algorithm on the recommended aromatherapy formula to obtain a plurality of first candidate aromatherapy formulas;

[0092] Filter and select the first candidate aromatherapy formula based on user needs to obtain a plurality of second candidate aromatherapy formulas.

[0093] The aromatherapy formula trend can reflect the current market trend, and the formula of other users of the same user type can represent the potential aromatherapy formula corresponding to the current user's demand. The aromatherapy formula trend is used to screen the formula of other users of the same user type, and is recommended to the user. The behavior of the similar user group is used to filter the potential aromatherapy selection, so as to ensure that the recommendation system not only considers the preference of the current user, but also considers the group trend of the user type, greatly improving the accuracy and relevance of the recommendation.

[0094] Using the collaborative filtering algorithm, a plurality of candidate formulas corresponding to the recommended formula are generated according to the preferences of similar users. When similar users are found, the algorithm will give priority to user groups with similar characteristics. For example, formula 1: front note: lemon 30%, middle note: rose 50%, and back note: amber 20%; formula 2: front note: mint 25%, middle note: ocean 55%, and back note: vanilla 20%; formula 3: front note: orange 35%, middle note: lavender 45%, and back note: sandalwood 20%, as the first candidate aromatherapy formula. However, these formulas may only be screened based on preference similarity and user type, and may not meet the specific needs of each user. Therefore, the first candidate solution needs to be screened in combination with user demand, and the formula that does not meet the demand is removed, and only the aromatherapy formula that meets the demand is retained to obtain the second candidate aromatherapy formula. For example, if a user prefers strong woody fragrance and is a business user, even if the collaborative filtering recommends a fresh floral aromatherapy formula, this aromatherapy formula will be screened out because it does not meet the user's demand. If the user needs a summer suitable perfume, the formula containing fresh fragrance is preferred. Through the collaborative filtering algorithm and user type analysis, accurate recommendation can be made according to the user's historical behavior and the preferences of similar users. When the user's demand is less common, collaborative filtering can effectively recommend through the behavior of similar users, which can alleviate the common data sparsity problem in recommendation operation.

[0095] Preferably, the aromatherapy formula trend can reflect the current market trend, and the formula of other users of the same user type can represent the potential aromatherapy formula corresponding to the current user's demand. The aromatherapy formula trend is used to screen the formula of other users of the same user type, and is recommended to the user. The behavior of the similar user group is used to filter the potential aromatherapy selection, so as to ensure that the recommendation system not only considers the preference of the current user, but also considers the group trend of the user type, greatly improving the accuracy and relevance of the recommendation.

[0096] Specifically, it is also proposed to assign a unique number to each fragrance to form a raw material database, which should contain the name, characteristics, odor description, usage suggestions and interactions of each fragrance, etc. The raw materials can be classified by type (plant extracts, synthetic fragrances, natural fragrances, etc.) and labeled (such as top, middle, and back) to facilitate selection during subsequent fragrance blending. When the user selects a set of raw materials for blending, a combination code is generated according to the number and amount of each fragrance. This combination code not only indicates the type of fragrance used, but also reflects the proportion of each fragrance in the formula. For example, using the raw material number and its corresponding component, a format similar to "raw material number. amount" is formed to facilitate quick identification and analysis. For example, if three fragrances are selected, the code is "001.30-002.20-003.50", indicating that the fragrance with number 001 is used in an amount of 30%, the fragrance with number 002 is used in an amount of 20%, and the fragrance with number 003 is used in an amount of 50%. This can facilitate robot recognition.

[0097] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "exemplary embodiment", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the exemplary description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0098] Although embodiments of the present application have been shown and described, those skilled in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the claims and their equivalents.

Claims

1. A three-dimensional machine vision based perfumery dual arm robot system, characterized in that, The aroma blending double-arm robot is configured with a three-dimensional machine vision module, an aroma blending algorithm module, and a mechanical arm provided with a gripper; The aroma blending algorithm module is configured to obtain aroma requirements of a user, generate a plurality of aroma recipes based on characteristics of aroma raw materials and the aroma requirements of the user, and provide the user with the aroma recipes for selection; The three-dimensional machine vision module is configured to identify and locate positions of the required aroma raw materials and aroma blending containers after the user selects at least one aroma recipe, and generate aroma blending instructions for driving the mechanical arm and the gripper; The aroma blending instructions are configured to drive the mechanical arm and the gripper to perform weighing, material adding, liquid transferring, and stirring operations according to the positions of the required aroma raw materials and the aroma blending containers, aroma recipe setting proportions, and aroma blending sequences, so as to obtain blended aromas, bottle and package the blended aromas, and output the blended aromas to the user; In the aroma blending algorithm module, generating a plurality of aroma recipes based on characteristics of aroma raw materials and aroma requirements of a user includes: Converting the aroma requirements of the user into a feature vector; Using clustering analysis and a decision tree to establish a user preference model and determining a user type based on the feature vector; Generating preselected aroma recipes according to the characteristics of the raw materials and the feature vector, and generating a plurality of alternative aroma recipes according to the user type using a collaborative filtering scheme; Iteratively optimizing the preselected aroma recipes and the alternative aroma recipes to obtain a plurality of preferred aroma recipes; In the aroma blending algorithm module, using clustering analysis and a decision tree to establish a user preference model and determining a user type based on a feature vector includes: Using clustering analysis to divide users with similar preferences into different types; Analyzing feature mean values of each user type to obtain common preferences of each group; Inputting the feature vector as a user feature into a decision tree to analyze the importance of the user feature; Combining the user type and the result of the importance analysis to obtain an aroma recipe trend; In the aroma blending algorithm module, generating preselected aroma recipes according to the characteristics of the raw materials and the feature vector includes: Obtaining characteristics of all raw materials; Obtaining a raw material combination based on the characteristics of the raw materials and aroma notes required by the user; Adjusting a conventional blending ratio according to a concentration requirement of the user to obtain a latest concentration ratio; Combining the raw material combination and the latest concentration ratio to obtain preselected aroma recipes.

2. The perfumed two-armed robotic system of claim 1, wherein, The aroma blending algorithm module is configured to receive aroma requirements input by a user, generate a plurality of aroma recipes, and visually output the aroma recipes to the user for confirmation; The aroma requirements include aroma notes, concentration, odor types, use occasions, seasons, and duration.

3. The perfumed two-armed robotic system of claim 1, wherein, The three-dimensional machine vision module is configured to pre-paste labels on aroma raw materials and aroma blending containers, find corresponding labels according to aroma recipes selected by a user to determine positions of required aroma raw materials and aroma blending containers after the user selects at least one aroma recipe, and generate aroma blending instructions for driving a mechanical arm and a gripper; After a user provides feedback on a recipe, incremental learning is performed based on evaluations of the user to optimize generation of a next aroma recipe.

4. The perfumed two-armed robotic system of claim 3, wherein, After the mechanical arm and the clamping piece receive the blending instruction, the position of the mechanical arm moving the clamping piece changes according to the blending sequence, so that the clamping piece moves to the position of the required aromatherapy raw material, the clamping piece takes the required aromatherapy raw material according to the set proportion of the aromatherapy formula, weighs the taken aromatherapy raw material, and then adds the weighed aromatherapy raw material into the blending container until all the required aromatherapy raw materials are added into the blending container according to the set proportion of the aromatherapy formula; The clamping piece takes the stirring tool to stir the aromatherapy raw material in the blending container to obtain the blended aromatherapy, and then adds the blended aromatherapy into different aromatherapy bottles according to the net content of the selected aromatherapy by the user, and reminds the user to take it.

5. The perfumed two-armed robotic system of claim 1, wherein, In the blending algorithm module, a plurality of candidate aromatherapy formulas are generated according to user types using a collaborative filtering scheme, including: Combining the formula selection and the aromatherapy formula trend of other users of the same user type, the aromatherapy formula is recommended to the user; Combining the user type and using the collaborative filtering algorithm on the recommended aromatherapy formula, a plurality of first candidate aromatherapy formulas are obtained; Filtering and screening the first candidate aromatherapy formula based on the user demand, a plurality of second candidate aromatherapy formulas are obtained.

6. The perfumed two-armed robotic system of claim 1, wherein, The blending instruction includes an aromatherapy combination code, and the aromatherapy combination code includes the required aromatherapy raw material and the set proportion of the aromatherapy formula.

Images