Full-automatic cooking control method and cooking machine

Abstract

The invention discloses a full-automatic cooking control method and a cooking machine. The full-automatic cooking control method comprises the steps that 1, a menu text is analyzed, actions and food material information are extracted, and a standardized food material list and an action sequence are generated; 2, calculating a temperature curve and a time sequence in combination with nutrition attenuation characteristics of the food materials; 3, controlling the feeding opportunity and the stir-frying rotating speed according to the temperature curve and the time sequence; step 4, calculating a nutrition score, and optimizing the seasoning amount by using the nutrition score and the seasoning cost; 5, quantifying the nutritional status of the dishes in combination with the seasoning amount, and outputting an evaluation result according to the user satisfaction; step 6, adjusting cooking parameters according to an evaluation result, and feeding back to the step 1 to realize taste individuation; according to the intelligent kitchen intelligent cooking system, the subject technologies of natural language processing, computer vision, thermodynamics, biology and reinforcement learning are fused, automation and personalized optimization of the whole cooking process are achieved, the consistency of cooking quality is remarkably improved, and an efficient and healthy cooking solution is provided for a household intelligent kitchen.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of intelligent cooking equipment, and particularly relates to a full-automatic frying control method and a frying machine. BACKGROUND

[0002] Traditional cooking relies on manual operation, which has many problems such as too much manual intervention, large fluctuation of dish quality caused by cooking experience difference, easy loss of nutrition, and low cooking efficiency. The existing automatic frying machine mostly adopts fixed program control, lacks precise analysis of a menu, intelligent identification of a food material state, and dynamic optimization of a cooking process, is difficult to adapt to diversified food materials and personalized cooking demands, and has defects such as high energy consumption and poor nutrition retention effect. Therefore, a frying method and equipment that integrates multidisciplinary technologies and realizes full-process automation and personalized optimization are urgently needed. SUMMARY

[0003] (1) Technical problem to be solved The application discloses a full-automatic frying control method and a frying machine, and aims to solve the problems of traditional cooking, such as too much manual intervention, large fluctuation of dish quality, and unbalanced nutrition.

[0004] (2) Technical scheme The application discloses a full-automatic frying control method, which comprises the following steps: Step 1, menu analysis and food material identification: analyzing a menu text to extract action and food material information, identifying actual food material types and states, and generating a standardized food material list and an action sequence; Step 2, heat conduction optimization: based on the food material information and the action sequence, in combination with food material nutrition attenuation characteristics, calculating an optimized temperature curve and a time sequence; Step 3, feeding and frying control: according to the temperature curve, the time sequence and the food material list, controlling feeding time and frying rotation speed, and recording an execution log; Step 4, seasoning and nutrition balance: based on the execution log and the food material list, calculating a nutrition score of a dish, using the nutrition score and a seasoning cost to optimize a seasoning amount, and ensuring nutrition balance; Step 5, dish taking out and feedback evaluation: judging a dish taking out time, quantifying dish nutrition status in combination with the seasoning amount, generating a quality evaluation in combination with user satisfaction, and outputting an evaluation result; Step 6, closed-loop optimization: adjusting cooking parameters according to the evaluation result, and feeding back to Step 1 to realize personalized optimization of dish taste.

[0005] Further, the step 1 comprises the following steps: Step 101, generating a recommended menu text according to a preset menu selected by a user or an actual food material image uploaded by the user, extracting corresponding food materials and food material associated information, and then extracting key actions through an action priority scoring rule; Step 102, using the multi-state, multi-angle synthetic image generated by Stable Diffusion to train the BLIP-2 model, combining the food material-state-semantic association graph to correct the semantics and optimize the recognition confidence formula, and then verifying the food material type and state of the actual food material through the optimized recognition confidence formula.

[0006] Further, the action priority scoring rule is to evaluate the priority of the action by combining the action frequency and the matching degree of the action and the food material from the Recipe1M dataset, and the specific formula is as follows:

[0007] wherein, represents the priority of the first step action, represents the priority of the first step cooking action, represents the priority of the second step action, represents the priority of the second step cooking action, represents the action frequency, represents the number of actions, represents the matching degree of the action and the food material list, represents the priority weight.

[0008] Further, the step 102 includes the following steps: Step 10201, constructing a knowledge graph , the node represents the food material list, the edge represents the relationship and state change between food materials, and the semantic consistency score of each food material is calculated through the GNN graph embedding algorithm, Step 10202, dynamically adjusting the temperature parameter of the model when generating the synthetic image , dynamically optimizing the image quality and diversity, and the formula is as follows:

[0009] wherein, represents the initial temperature, represents the temperature decay rate, represents the current training round, represents the maximum training round; Step 10203, combining the image feature distance , the semantic consistency score and the temperature adjustment factor and the temperature adjustment factor , optimizing the recognition confidence formula, and the optimization formula is as follows:

[0010] wherein, represents the priority of the first step action, ​Food The confidence level of identification, Indicates the first Food Image feature distance, Indicates the first Food The semantic consistency score of the knowledge graph. Indicates the temperature regulation factor. Indicates the feature distance scaling factor. Represents semantic consistency weight. Indicates temperature regulation weight. Indicates the confidence threshold; Temperature regulation factor The formula is:

[0011] in, Indicates temperature regulation sensitivity. Indicates the critical temperature value. Indicates temperature parameter; Step 10204: Use the recognition confidence level to determine and verify the type and state of the ingredients. When the confidence level is lower than the recognition threshold, remind the user to adjust the ingredients; otherwise, proceed to the next step.

[0012] Furthermore, step 2 is specifically as follows: Step 201: Using the ingredient list and action sequence, construct a multi-agent optimization framework containing multiple agents. The agents include energy consumption, nutrient retention rate and taste score, so that the framework can simultaneously take into account multiple optimization goals such as low energy consumption, high nutrient retention and good taste. Step 202: Introduce the synergy of multiple heat sources and the dynamic decay rate of ingredients to construct a temperature curve, so as to dynamically simulate the relationship between temperature changes and nutrient loss during cooking. The formula for the temperature curve is as follows:

[0013] The dynamic decay rate formula is as follows:

[0014] in, Indicates the current time instantaneous temperature below Indicates the time from start to current time. At any time in, Indicates time instantaneous temperature below Indicates the initial temperature. Indicates the first The conductivity coefficient of a heat source, represents the first heat source function, represents the first heat source function, represents the dynamic nutritional decay rate of the first heat source, represents the first heat source function, represents the first heat source function, represents the first heat source function, represents the temperature sensitivity coefficient, represents the first heat source function, represents the first heat source function, represents the number of heat sources; Step 203, using heat diffusion strategy, the temperature control in the temperature curve is converted into heat flow tracking in the parameter space, guiding the agent to find the globally optimal temperature curve by tracking the heat flow direction; Step 204, using the optimized temperature curve and combining multiple optimization objectives of low energy consumption, high nutritional retention and good taste, the optimal cooking time sequence is calculated.

[0015] Further, the step 3 is specifically as follows: Step 301, using the temperature curve, time sequence and food material list, the feeding device is controlled to feed the food materials according to the feeding time rule in order of the food material maturity; wherein, the feeding time rule is to accurately calculate the feeding time of each food material in the cooking action based on the total time length of the corresponding cooking action, combined with the maturity proportion of the food material already fed and all food materials to be fed; Step 302, the stir-frying device adjusts the real-time rotating speed based on the reference rotating speed, combined with the periodic fluctuation control function and the intelligent force adjusting mechanism, and adjusts the rotating speed fluctuation amplitude according to the real-time density change of the food material, Step 303, recording the execution log.

[0016] Further, the step 4 is specifically as follows: Step 401, the seasoning and nutritional balance module receives the execution log and the food material list, and evaluates the dish nutritional score under multiple seasonings through the nutritional score calculation rule; Step 402, taking the high nutritional score and low seasoning cost of the dish as the target, the optimal seasoning amount is sought, which includes the input amount and corresponding input time of multiple seasonings, so as to accurately quantitatively feed the seasonings, dynamically improve the nutritional retention rate, and realize nutritional balance.

[0017] Further, the step 5 is specifically as follows: Step 501, out of the pot judgment: using the difference between the real-time temperature collected by the temperature sensor and the target temperature as an exponential term, and combining the sensitivity factor to calculate the out of the pot probability, when reaching the preset threshold, the frying machine is controlled to complete the out of the pot action; Step 502, quality evaluation: the nutritional score of the dish is calculated by using the optimized seasoning amount, and the evaluation result is generated by combining the user input satisfaction score.

[0018] Further, the step 6 is specifically as follows: Step 601, the user behavior change amount is calculated by using the evaluation result, and the product of the behavior change amount and the system efficiency coefficient is obtained to obtain the influence intensity, Step 602, on the basis of the initial parameters, the product of the influence intensity and the learning rate is combined to optimize and generate the cooking parameters, and the cooking parameters are fed back to step 1 to adjust the food material list and the action sequence.

[0019] The application discloses a kind of fried machine, including recipe analysis and food material identification module, heat conduction optimization module, feeding and stir-frying control module, seasoning and nutrition balance module, out of pot and feedback learning module, user interaction and influence module and control processor, the control processor is connected with each described module respectively, to make the fried machine execute as above described whole automatic stir-frying control method.

[0020] Compared with prior art, the beneficial effects of the present application are: The present application realizes the whole process automation and individual optimization of cooking by six steps of recipe analysis and food material identification, heat conduction optimization, feeding and stir-frying control, seasoning and nutrition balance, out of pot and feedback evaluation and closed-loop optimization, and combines natural language processing, computer vision, thermodynamics, biology and reinforcement learning and other disciplines and technologies, so that the consistency of cooking quality is significantly improved, and an efficient and healthy cooking solution for family intelligent kitchen is provided. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The overall flowchart of the implementation method of the present application is shown.

[0022] Figure 2 The flowchart of step 1 of the present application is shown.

[0023] Figure 3 The flowchart of step 102 of the present application is shown.

[0024] Figure 4 The flowchart of step 2 of the present application is shown. DETAILED DESCRIPTION

[0025] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work belong to the scope of protection of the present application. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0026] In order to better understand the technical content of the present embodiment, the terms involved in the present embodiment will be explained first.

[0027] NLP: Natural Language Processing, aims to enable computers to understand and process human language, extract useful information from it, and help humans more efficiently handle various tasks.

[0028] Stable Diffusion: an open-source text-to-image deep learning model, the core function is to generate corresponding images according to text descriptions, also supports image editing, style transfer, etc.

[0029] Recipe1M dataset: a large-scale multi-modal recipe dataset built by MIT team, the core is used for food-related artificial intelligence research, and is one of the most representative public datasets in the food field, containing more than 100,000 recipes and about 800,000-1.3 million corresponding food images.

[0030] GNN graph embedding algorithm: a technology that combines graph neural networks (GNN) with graph embedding, the core is to convert nodes (or edges) in the graph into low-dimensional vectors (embedding), while preserving the structure, attributes and semantic information of the graph.

[0031] Figure 1 The flowchart of the embodiment of the present application is shown, and the present application discloses a full-automatic cooking control method, comprising the following steps: Step 1, recipe analysis and ingredient recognition: analyze the recipe text to extract action and ingredient information, recognize the actual ingredient type and state, and generate a standardized ingredient list and action sequence; Step 2, heat conduction optimization: based on the ingredient information and action sequence, combined with the nutrient decay characteristics of the ingredients, calculate the optimized temperature curve and time sequence; Step 3, feeding and stir-frying control: according to the temperature curve, time sequence and ingredient list, control the feeding time and stir-frying speed, and record the execution log; Step 4, seasoning and nutrient balance: based on the execution log and ingredient list, calculate the nutrient score of the dish, and use the nutrient score and seasoning cost to optimize the seasoning amount to ensure nutrient balance; Step 5, dish out and feedback evaluation: determine the dish out timing, quantify the nutritional status of the dish combined with the seasoning amount, generate a quality evaluation with user satisfaction, and output the evaluation results; Step 6, closed-loop optimization: adjust the cooking parameters according to the evaluation results, feedback to step 1, and realize personalized optimization of dish taste.

[0032] The implementation process of the full-automatic frying control method will be described in detail below in conjunction with an embodiment: Perform step 1, recipe analysis and ingredient recognition: The recipe analysis and ingredient recognition module includes a recipe NLP analysis module and an ingredient image recognition module, which are used to analyze the recipe text to extract action and ingredient information, recognize the actual ingredient type and state, and generate a standardized ingredient list and action sequence , wherein, represents the th ingredient and its associated information (including ingredient density, state, and ease of cooking, etc.), represents the th cooking action; specifically including the following steps: Step 101, the user selects a preset recipe or uploads an actual ingredient image to the recipe NLP analysis module to generate a corresponding recommended recipe text, the recipe NLP analysis module uses a fine-tuning NLP model based on BLIP and extracts ingredients and their associated information according to the recipe text content, and then extracts key actions through action priority scoring rules, combined with action frequency statistics from Recipe1M dataset and the matching degree of the action and the ingredient. Further, the action priority scoring rule formula is as follows: ;

[0033] , wherein, represents the th action priority, represents the th cooking action, represents the action frequency (obtained from Recipe1M dataset), represents the number of actions, represents the matching degree of the action and the ingredient list (taking a value between 0 and 1), represents the priority weight; Step 102: The food image recognition module uses multi-state, multi-angle synthetic images generated by Stable Diffusion to train the BLIP-2 model. It then performs semantic correction and optimizes the recognition confidence formula by combining the food-state-semantic association graph. The optimized recognition confidence formula is then used to verify the food type and state of the uploaded food image. When the recognition confidence is lower than the recognition threshold, the system prompts the user to adjust the food; otherwise, it proceeds to the next cooking step. Specifically, the steps include: Step 10201: Construct a knowledge graph ,node Represents a list of ingredients, with sides It represents the relationships between ingredients (such as "pairing" and "substitution") and state changes (such as "raw → cooked" and "block → slice"). It calculates the semantic consistency score of each ingredient through the GNN graph embedding algorithm to correct the image recognition results. It uses knowledge graph to impose semantic constraints on the image recognition results to solve the problem of misrecognition by traditional models in complex scenes (occlusion, shadow, blur, etc.). Step 10202: Dynamically adjust the temperature parameters of the model when generating the synthesized image. The model dynamically optimizes the quality and diversity of generated images to improve its robustness, as shown in the following formula:

[0034] in, Indicates the initial temperature. Indicates the temperature decay rate. Indicates the current training round. Indicates the maximum number of training rounds; Step 10203, integrate image feature distance Semantic consistency score and temperature regulation factor The recognition confidence formula was optimized to achieve higher accuracy and adaptability. The formula for confidence level identification is as follows:

[0035] in, Indicates the first Food The confidence level of identification, Indicates the first Food Image feature distance, Indicates the first Food The semantic consistency score of the knowledge graph. Indicates the temperature regulation factor. Indicates the feature distance scaling factor. represents a semantic consistency weight, represents a temperature adjustment weight, represents a confidence threshold; temperature adjustment factor The formula is:

[0036] wherein, represents a temperature adjustment sensitivity, represents a temperature threshold, represents a temperature parameter; Step 10204, the type and state of the food material are judged and verified by using the recognition confidence, when the confidence is lower than the recognition threshold, the user is reminded to adjust the food material, otherwise the next step operation is entered; By dynamically adjusting the temperature parameter of the Stable Diffusion generated image, the diversity and quality trade-off of the generated sample are controlled by using the temperature adjustment idea; lower temperature can focus on high probability food material features (improve accuracy), while higher temperature can increase diversity (deal with complex scenes); At the same time, by respectively using the action priority scoring rule and the optimized recognition confidence, a standardized food material list and step sequence closer to the actual food material are generated, providing data basis for subsequent optimization, ensuring that the system understands the recipe intention and avoids ambiguity analysis errors, so as to accurately control the dish.

[0037] Step 2, heat conduction optimization: the heat conduction optimization module calculates the optimized temperature curve and time sequence based on the food material information and action sequence, combined with the food material nutrition attenuation characteristics, which includes the following steps: Step 201, the heat conduction optimization module receives the food material list and action sequence output by step 1, and constructs a multi-agent optimization framework containing multiple agents, including energy consumption, nutrition retention rate and taste score, so that the framework simultaneously considers multiple optimization objectives of less energy consumption, more nutrition retention and good taste; Step 202, introduce multi-heat source cooperation and dynamic attenuation rate of food material to construct temperature curve, to dynamically simulate the relationship between temperature change and nutrition loss in cooking; The temperature curve formula is as follows:

[0038] The dynamic attenuation rate formula is as follows:

[0039] wherein, represents the instantaneous temperature at the current time represents the instantaneous temperature at the current time represents the instantaneous temperature at the current time any of the time points, representing the instantaneous temperature at the time point , representing the initial temperature, representing the conduction coefficient of the th heat source, representing the th heat source function, representing the dynamic nutritional decay rate of the th heat source, representing the reference decay rate of the th heat source, representing the temperature sensitivity coefficient, representing the critical temperature of the th heat source, representing the number of heat sources; by modeling the dynamic decay rate based on enzyme activity and multi-heat source synergy, the temperature curve is self-adaptive to accurately simulate complex cooking environments; Step 203, using heat diffusion strategy, convert temperature control in temperature curve to heat flow tracking in parameter space, guide agent to find global optimal temperature curve by tracking heat flow direction; this step combines heat diffusion optimization with multi-agent system to dynamically balance multiple objectives such as energy, nutrition and taste, and find the global optimal temperature curve; Step 204, using the optimized temperature curve and combining multiple optimization objectives such as low energy consumption, high nutrition retention and good taste, calculate the optimal cooking time sequence; In the heat conduction optimization module, by more accurately calculating the cooking temperature and cooking time, the energy consumption is minimized while the nutrition retention and taste performance are maximized, ensuring an efficient, healthy and user-desired cooking process.

[0040] Step 3, feeding and stir-frying control: the feeding and stir-frying control module controls the feeding time and stir-frying speed according to the temperature curve, time sequence and food list, and records the execution log; specifically including the following steps: Step 301, after receiving the temperature curve, time sequence and food list, the feeding and stir-frying control module controls the feeding device to feed the food according to the feeding time rule based on the food maturity; for example, when cooking green pepper and potato shreds, since potato has high maturity and green pepper has low maturity, potato needs to be put into the pot first, and after a period of cooking, green pepper is put into the pot to ensure that both reach the best taste at the same time; specifically, the feeding time rule is to take the total duration of the corresponding cooking action as the basis, and combine the maturity proportion of the food already fed and the total food to be fed to accurately calculate the feeding time of each food in the cooking action; Step 302, the feeding and stir-frying control module controls the stir-frying device to combine the periodic fluctuation control function ( The function and intelligent force adjustment mechanism adjust the real-time speed, so that the stir-frying device periodically simulates the "push-pull-shake" rotating stir-frying action. At the same time, the speed fluctuation range is adjusted according to the real-time density change of the ingredients. When the density of the ingredients is high during the raw stage, the speed fluctuation range is large (large-scale stir-frying); when the ingredients are close to being cooked and the density decreases, the speed fluctuation range is reduced, so as to avoid the ingredients being broken. Step 303: Record the execution log.

[0041] Step 4, Seasoning and Nutritional Balance: The seasoning and nutritional balance module calculates the nutritional score of the dish based on the execution log and ingredient list. It then optimizes the seasoning amount using the nutritional score and seasoning cost to ensure nutritional balance. Specifically, this includes the following steps: Step 401: The seasoning and nutrition balance module receives the execution log and ingredient list, and evaluates the nutritional score of dishes under various seasonings according to the nutritional score calculation rules. Furthermore, the rule for calculating the nutritional score of a dish is to first calculate the score of each individual nutrient, and then sum up the scores of each nutrient to obtain the overall nutritional score of the dish. The specific formula is as follows:

[0042] in, This represents the scores of individual nutrients under different nutrient weights. This indicates the effect of various combinations of input amounts and input timings on the retention of this nutrient by different seasonings. Content of various nutrients Indicates the first Maximum recommended daily intake of 100 nutrients Indicates saturation factor. It is the first Nutritional weight of each nutrient Indicates the total category of nutrients. Indicates the nutritional score of the dish; It should be noted that because seasonings themselves contain certain nutrients, and the timing of their addition can also affect the nutrient retention of a dish (for example, adding salt too early when stir-frying will cause the food to release water, resulting in some sodium loss with the broth, reducing the sodium retention of the finished dish; while adding salt just before serving ensures that almost all the sodium is retained in the dish; therefore, the amount and timing of salt addition affect the sodium retention of the dish). It includes the nutrient content obtained under different combinations of seasoning dosage and timing, which facilitates the optimization of subsequent seasoning dosage; Step 402, then with high nutritional score of dishes, low seasoning cost as the target, seek the optimal seasoning amount, the seasoning amount contains the input amount of multiple seasonings and the corresponding input timing, in order to reasonably control the seasoning device to accurately quantify the seasoning, to dynamically improve the nutrient retention rate, achieve nutritional balance, adapt to the user's health needs.

[0043] Step 5, out of the pot and feedback evaluation: The out of the pot and feedback learning module judges the out of the pot opportunity, combines the seasoning amount to quantify the nutritional status of the dish and generates a quality evaluation with the user satisfaction, and outputs the evaluation result; Specifically, the following steps are included: Step 501, out of the pot judgment: The out of the pot and feedback learning module uses the difference between the real-time temperature collected by the temperature sensor and the target temperature as an exponential term, and then combines the sensitivity factor to calculate the out of the pot probability. When reaching the preset threshold, control the frying machine to complete the out of the pot action; Wherein, the sensitivity factor is used to control the sensitivity change of the out of the pot probability to the temperature change, so as to accurately control the out of the pot time; Step 502, quality evaluation: Calculate the nutritional score of the dish using the optimized seasoning amount, and generate an evaluation result combined with the user's satisfaction score; In this step, by combining "objective calculation of nutritional data" with "subjective user feedback", the comprehensive quality score is obtained, which provides a quantitative basis for subsequent cooking parameter optimization.

[0044] Step 6, closed-loop optimization: The user interaction and influence module adjusts the cooking parameters according to the evaluation result and feeds back to step 1 to realize personalized optimization of dish taste; Specifically, the following steps are included: Step 601, the user interaction and influence module receives the evaluation result and calculates the user behavior change amount, and calculates the influence intensity using the product of the behavior change amount and the system efficiency coefficient. The influence intensity represents the degree of influence of user behavior change on the model. When the influence intensity is large, it means that the dish does not meet the user's taste and the weight parameters of the model need to be adjusted significantly. When the influence intensity is small, it means that the dish meets the user's taste and only needs to fine-tune the weight parameters of the model or does not need to be adjusted. Step 602, on the basis of the initial parameters, combine the product of the influence intensity and the learning rate to optimize and generate the cooking parameters, and then feed back the cooking parameters to the recipe analysis and ingredient recognition module in step 1 to further adjust the ingredient list and action sequence according to the user's taste, so that the frying machine gradually adapts to the user's taste preference; Wherein, the learning rate is used to control the adjustment amplitude of the parameters, to avoid too large adjustment in a single time, so that the dish taste changes too much beyond the expectation; In this step, the user feedback in step 5 is quantified into an influence value, and then the cooking parameters are fine-tuned based on the learning rate, so that the frying machine can make dishes that meet the user's taste better next time, thereby realizing personalized dish taste optimization.

[0045] The embodiment of the present application also provides a cooking machine, which comprises a recipe analysis and food material recognition module, a heat conduction optimization module, a feeding and stir-frying control module, a seasoning and nutrition balance module, a dish serving and feedback learning module, a user interaction and influence module and a control processor, the control processor is connected with each of the modules, so that the cooking machine performs the automatic stir-frying control method as described above, and the specific working principle can refer to the corresponding process in the foregoing method embodiment, which will not be described here in detail.

[0046] The experimental environment of the present application includes a hardware environment: the image recognition experiment is run on an NVIDIA RTX 3090 GPU (24 GB video memory), and the heat conduction model optimization uses an Intel Xeon Gold 6226R CPU (16 cores) for multi-agent reinforcement learning calculation. The cooking machine prototype is equipped with a high-precision temperature sensor (±0.1°C), a speed controller (±1 RPM) and an automatic feeding device.

[0047] A software environment: the image recognition module is implemented based on PyTorch 1.9.0, and uses BLIP-2 model and StableDiffusion; the heat conduction optimization module uses TensorFlow 2.5.0 to implement multi-agent reinforcement learning (DQN algorithm); the knowledge graph is constructed based on a Neo4j database.

[0048] The present application realizes the automatic and personalized optimization of the whole cooking process through six steps of recipe analysis and food material recognition, heat conduction optimization, feeding and stir-frying control, seasoning and nutrition balance, dish serving and feedback evaluation and closed-loop optimization, and combines natural language processing, computer vision, thermodynamics, biology and reinforcement learning and other disciplines and technologies, so that the consistency of cooking quality is significantly improved, and an efficient and healthy cooking solution is provided for a family smart kitchen.

[0049] The technical features or steps of the above embodiments can be combined in any way. To make the description concise, not all possible combinations of the technical features or steps in the above embodiments are described, but as long as the combinations of the technical features do not exist, they should be considered as the scope of the present application.

[0050] Those skilled in the art will appreciate that embodiments of the application can be provided as methods, systems or computer program products. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

Claims

1. A fully automatic cooking control method, characterized by, The method comprises the following steps: Step 1, recipe analysis and ingredient recognition: analyze the recipe text to extract action and ingredient information, identify the actual ingredient type and state, and generate a standardized ingredient list and action sequence; Step 2, heat conduction optimization: based on the ingredient information and action sequence, combined with the nutrient decay characteristics of the ingredients, calculate the optimized temperature curve and time sequence; Step 3, feeding and stir-frying control: according to the temperature curve, time sequence and ingredient list, control the feeding time and stir-frying speed, and record the execution log; Step 4, seasoning and nutritional balance: based on the execution log and ingredient list, calculate the nutritional score of the dish, use the nutritional score and seasoning cost to optimize the seasoning amount, and ensure nutritional balance; Step 5, out of the pot and feedback evaluation: determine the out of the pot opportunity, combine the seasoning amount to quantify the nutritional status of the dish and generate quality evaluation with user satisfaction, output the evaluation results; Step 6, closed-loop optimization: adjust the cooking parameters according to the evaluation results and feed back to step 1 to realize personalized optimization of dish taste.

2. The full-automatic cooking control method according to claim 1, characterized in that, The step 1 comprises the following steps: Step 101, generate recommended recipe text according to user-selected preset recipe or uploaded actual ingredient image, and extract corresponding ingredients and ingredient-related information, then extract key actions through action priority scoring rules; Step 102, use the multi-state and multi-angle synthetic images generated by Stable Diffusion to train the BLIP-2 model, combine the ingredient-state-semantic association graph for semantic correction and optimize the recognition confidence formula, and then verify the ingredient type and state of the actual ingredients through the optimized recognition confidence formula.

3. The full-automatic cooking control method according to claim 2, characterized in that, The action priority scoring rule is to evaluate the priority of the action by combining the action frequency and the matching degree of the action and the ingredient from the Recipe1M dataset, and the specific formula is as follows: ； in, Indicates the first Step action priority, Indicates the first Cooking steps, Indicates the frequency of action. Indicates the number of actions. Indicates a list of actions and ingredients. The degree of matching, Indicates priority weight.

4. The full-automatic cooking control method according to claim 2, characterized in that, The step 102 comprises the following steps: Step 10201, construct a knowledge graph , node represent a list of ingredients, edge represent the relationship between ingredients and state changes, calculate the semantic consistency score of each ingredient by GNN graph embedding algorithm, Step 10202, dynamically adjusting the temperature parameter of the model when generating the synthetic image , dynamically optimizing the quality and diversity of the generated image, the formula is as follows: ； wherein, denotes the initial temperature, denotes the temperature decay rate, denotes the current training epoch, denotes the maximum training epoch; Step 10203, integrate image feature distance , semantic consistency score and temperature adjustment factor Optimize the recognition confidence formula, the optimization formula is as follows: ； wherein, represents the recognition confidence of the first food material, represents the image feature distance of the first food material, represents the knowledge graph semantic consistency score of the first food material, represents the temperature adjustment factor, represents the feature distance scaling factor, represents the semantic consistency weight, represents the temperature adjustment weight, represents the confidence threshold value;​​​​​​ Temperature regulating factor The formula is: ； wherein denotes the temperature regulation sensitivity, denotes the temperature threshold value, denotes the temperature parameter; Step 10204, use the recognition confidence to verify the ingredient type and state, if the confidence is lower than the recognition threshold, remind the user to adjust the ingredients, otherwise go to the next step.

5. The full-automatic cooking control method according to claim 1, characterized in that, The step 2 comprises the following steps: Step 201, use the ingredient list and action sequence to build a multi-agent optimization framework containing multiple agents, including energy consumption, nutrient retention rate and taste score, so that the framework can simultaneously consider multiple optimization objectives such as low energy consumption, high nutrient retention and good taste; Step 202, introduce multi-heat source cooperation and dynamic decay rate of ingredients to build temperature curve to dynamically simulate the relationship between temperature change and nutrient loss in cooking; The temperature curve formula is as follows: ； The dynamic decay rate formula is as follows: ； in, Indicates the current time instantaneous temperature below Indicates the time from start to current time. At any time in, Indicates time instantaneous temperature below Indicates the initial temperature. Indicates the first The conductivity coefficient of a heat source, Indicates the first A heat source function, express The dynamic nutrient decay rate of each heat source Indicates the first The reference decay rate of each heat source Indicates the temperature sensitivity coefficient. Indicates the first The critical temperature of a heat source. Indicates the number of heat sources; Step 203, use heat diffusion strategy to convert temperature control in temperature curve to heat flow tracking in parameter space, guide agents to find the globally optimal temperature curve by tracking heat flow direction; Step 204, use the optimized temperature curve and combine multiple optimization objectives such as low energy consumption, high nutrient retention and good taste to calculate the optimal cooking time sequence.

6. The full-automatic cooking control method according to claim 1, characterized in that, The step 3 comprises the following steps: Step 301, using the temperature curve, time series and food list, the feeding device is controlled to feed the food according to the feeding time rule according to the food maturity; wherein, the feeding time rule is to take the total length of the corresponding cooking action as the basis, and combine the maturity proportion of the food already fed and the total food to be fed to accurately calculate the feeding time of each food in the cooking action; Step 302, the stir-frying device adjusts the real-time speed based on the reference speed, combines the periodic fluctuation control function and the intelligent force adjusting mechanism, and adjusts the speed fluctuation amplitude according to the real-time density change of the food, Step 303, record the execution log.

7. The full-automatic cooking control method according to claim 1, characterized in that, The specific steps of step 4 are as follows: Step 401, the seasoning and nutrition balance module receives the execution log and the food list, and evaluates the dish nutrition score under multiple seasonings through the nutrition score calculation rule; Step 402, taking the high nutrition score and low seasoning cost of the dish as the target, the optimal seasoning amount is sought, which includes the input amount of multiple seasonings and the corresponding input time, to accurately quantify the seasoning and dynamically improve the nutrition retention rate, realizing nutrition balance.

8. The full-automatic cooking control method according to claim 1, characterized in that, The specific steps of step 5 are as follows: Step 501, out of the pot judgment: using the difference between the real-time temperature collected by the temperature sensor and the target temperature as the exponential term, and combining the sensitivity factor to calculate the out of the pot probability, when reaching the preset threshold, the frying machine is controlled to complete the out of the pot action; Step 502, quality evaluation: using the optimized seasoning amount to calculate the nutrition score of the dish, combining the user input satisfaction score to generate the evaluation result.

9. The full-automatic cooking control method according to claim 1, characterized in that, The specific steps of step 6 are as follows: Step 601, using the evaluation result to calculate the user behavior change amount, calculating the product of the behavior change amount and the system efficiency coefficient to obtain the influence strength, Step 602, on the basis of the initial parameters, combining the product of the influence strength and the learning rate to optimize and generate the cooking parameters, and then feeding the cooking parameters back to step 1 to adjust the food list and the action sequence.

10. A cooking machine, characterized by, The frying machine comprises a recipe analysis and food recognition module, a heat conduction optimization module, a feeding and stir-frying control module, a seasoning and nutrition balance module, an out of the pot and feedback learning module, a user interaction and influence module, and a control processor, wherein the control processor is connected with each module to enable the frying machine to perform the full-automatic frying control method as claimed in any one of claims 1-9.

Images