Intelligent cleaning method and terminal based on deep learning

Abstract

The present invention discloses an intelligent cleaning method and terminal based on deep learning, which determine the product type, debris state, fixture information and machine information of a test product; use deep learning technology to perform debris cleaning simulation on the test product according to the product type, debris state, fixture information and machine information to obtain a corresponding optimal blowing trajectory, and store the optimal blowing trajectory in a database; select a target blowing trajectory from the database according to the product type, debris state, fixture information and machine information of the product to be cleaned; clean the debris on the product to be cleaned according to the target blowing trajectory to obtain a cleaned product, use deep learning technology to learn different blowing trajectories according to different product types, debris states, fixture information and machine information, and subsequently select the blowing trajectory according to the information corresponding to the product to be cleaned for targeted cleaning, thereby efficiently cleaning the residual debris on the product.

Description

Technical Field

[0001] The present invention relates to the field of intelligent cleaning technology, and in particular to an intelligent cleaning method and terminal based on deep learning. Background Art

[0002] In a machining workshop, aluminum chips will remain on the surface of the product after the product is processed. In the prior art, manual air blowing or automatic air blowing devices are usually used to clean the chips. Manual air blowing to clean the chips is inefficient and requires a lot of labor costs. When using an automatic air blowing device to clean, since the automatic air blowing device has a fixed air blowing path for any product and chips, it still requires manual hand-held operation, which is inefficient. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide an intelligent cleaning method and terminal based on deep learning, which can efficiently clean residual debris on the product.

[0004] In order to solve the above technical problems, a technical solution adopted by the present invention is:

[0005] An intelligent cleaning method based on deep learning, comprising the steps of:

[0006] Determine the product type, chip status, fixture information, and machine information of the test product;

[0007] Using deep learning technology to perform debris cleaning simulation on the test product according to the product type, debris state, fixture information, and machine information, obtain the corresponding optimal blowing trajectory, and store the optimal blowing trajectory in a database;

[0008] Selecting a target blowing trajectory from the database according to the product type, debris status, fixture information, and machine information of the product to be cleaned;

[0009] The debris on the product to be cleaned is cleaned according to the target blowing trajectory to obtain a cleaned product.

[0010] In order to solve the above technical problems, another technical solution adopted by the present invention is:

[0011] A deep learning-based intelligent cleaning terminal includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are implemented:

[0012] Determine the product type, chip status, fixture information, and machine information of the test product;

[0013] Using deep learning technology to perform debris cleaning simulation on the test product according to the product type, debris state, fixture information, and machine information, obtain the corresponding optimal blowing trajectory, and store the optimal blowing trajectory in a database;

[0014] Selecting a target blowing trajectory from the database according to the product type, debris status, fixture information, and machine information of the product to be cleaned;

[0015] The debris on the product to be cleaned is cleaned according to the target blowing trajectory to obtain a cleaned product.

[0016] The beneficial effects of the present invention are as follows: deep learning technology is used to simulate the debris cleaning of the test product according to the product type, debris state, fixture information and machine information, and the corresponding optimal blowing trajectory is obtained, and the optimal blowing trajectory is stored in a database; a target blowing trajectory is selected from the database according to the product type, debris state, fixture information and machine information of the product to be cleaned, and the debris on the product to be cleaned is cleaned according to the target blowing trajectory to obtain a cleaned product. Manual blowing cleaning is no longer used as in the prior art, but deep learning technology is used to learn different blowing trajectories according to different product types, debris states, fixture information and machine information. Subsequently, the blowing trajectory can be selected according to the information corresponding to the product to be cleaned for targeted cleaning, thereby efficiently cleaning the residual debris on the product. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flowchart of a deep learning-based intelligent cleaning method according to an embodiment of the present invention;

[0018] Figure 2 This is a schematic diagram of the structure of a deep learning-based intelligent cleaning terminal according to an embodiment of the present invention;

[0019] Figure 3 This is a cleaning flow chart of the deep learning-based intelligent cleaning method in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to explain the technical content, achieved objectives and effects of the present invention in detail, the following is an explanation in combination with the implementation modes and the accompanying drawings.

[0021] Please refer to Figure 1 , an embodiment of the present invention provides an intelligent cleaning method based on deep learning, comprising the steps of:

[0022] Determine the product type, chip status, fixture information, and machine information of the test product;

[0023] Using deep learning technology to perform debris cleaning simulation on the test product according to the product type, debris state, fixture information, and machine information, obtain the corresponding optimal blowing trajectory, and store the optimal blowing trajectory in a database;

[0024] Selecting a target blowing trajectory from the database according to the product type, debris status, fixture information, and machine information of the product to be cleaned;

[0025] The debris on the product to be cleaned is cleaned according to the target blowing trajectory to obtain a cleaned product.

[0026] From the above description, it can be seen that the beneficial effects of the present invention are: using deep learning technology to simulate the debris cleaning of the test product according to the product type, debris state, fixture information and machine information, and obtain the corresponding optimal blowing trajectory, and store the optimal blowing trajectory in the database, and select the target blowing trajectory from the database according to the product type, debris state, fixture information and machine information of the product to be cleaned, and clean the debris on the product to be cleaned according to the target blowing trajectory to obtain the cleaned product. It is no longer like the prior art that uses manual blowing cleaning, but uses deep learning technology to learn different blowing trajectories according to different product types, debris states, fixture information and machine information. Subsequently, the blowing trajectory can be selected according to the information corresponding to the product to be cleaned for targeted cleaning, thereby efficiently cleaning the residual debris on the product.

[0027] Furthermore, the use of deep learning technology to perform debris cleaning simulation on the test product according to the product type, debris state, fixture information and machine information to obtain the corresponding optimal blowing trajectory includes:

[0028] Acquire a status image of the test product, wherein the status image includes a debris image;

[0029] Acquire a simulated blowing trajectory corresponding to the state image until the debris image does not exist in the state image;

[0030] Combining all the simulated blowing trajectories into an initial blowing trajectory;

[0031] Performing deep learning on the initial blowing trajectory according to the product type, debris state, fixture information, and machine information to obtain a learned blowing trajectory;

[0032] Cleaning debris from a preset number of inspection products according to the learned blowing trajectory to obtain a corresponding cleaning accuracy and cleaning time;

[0033] Determine whether the cleaning accuracy reaches a first preset value and whether the cleaning time is less than a second preset value. If so, determine the learned blowing trajectory as the optimal blowing trajectory. If not, return to execute the step of obtaining the status image of the test product.

[0034] From the above description, it can be seen that the debris state of each time and its corresponding simulated blowing trajectory as well as the debris state obtained after the simulated blowing trajectory is run are used as the learning set. When deep learning of the blowing trajectory is performed, the features in the simulated blowing trajectory are matched with the debris state according to the cleaning accuracy and cleaning time, and the optimal blowing trajectory is determined, which can accurately clean the debris on the product, thereby improving the cleaning quality of residual debris.

[0035] Furthermore, the determining of the product type, debris status, fixture information and machine information of the test product includes:

[0036] Collect a large number of different debris states to form a debris state library;

[0037] The selecting a target blowing trajectory from the database according to the product type, debris status, fixture information, and machine information of the product to be cleaned includes:

[0038] Determine the debris state of the product to be cleaned using visual inspection technology based on the debris state library;

[0039] Determine the product type, fixture information, and machine information of the product to be cleaned;

[0040] The target blowing trajectory is selected from the database according to the product type, debris status, fixture information and machine information of the product to be cleaned.

[0041] From the above description, it can be seen that by collecting a large number of different debris states to form a debris state library, the debris state of the product to be cleaned can be accurately determined by using visual inspection technology based on the debris state library, and the corresponding blowing trajectory can be used for cleaning according to different product types, debris states, fixture information and machine information, thereby improving the intelligence level of product debris cleaning.

[0042] Further, the cleaning of debris on the product to be cleaned according to the target blowing trajectory to obtain the cleaned product includes:

[0043] Cleaning the debris on the product to be cleaned according to the target blowing trajectory to obtain an initially cleaned product;

[0044] Using visual inspection technology to detect whether there are debris in the product after the initial cleaning, if not, it is determined that the cleaning is complete and the cleaned product is obtained, and if there are debris, the position corresponding to the debris that has not been cleaned is determined;

[0045] The product after the initial cleaning is cleaned by fixed-point directional blowing according to the positions corresponding to the debris that has not been cleaned up, until there is no debris in the product after the initial cleaning, so as to obtain a cleaned product.

[0046] From the above description, it can be seen that after obtaining the product after initial cleaning, visual inspection technology is used to detect whether there is any debris on the product after initial cleaning. If so, the position corresponding to the debris that has not been cleaned up is determined, and cleaning is performed again until there is no debris, thereby ensuring that the debris on the product is completely cleaned up.

[0047] Furthermore, storing the optimal blowing trajectory in a database includes:

[0048] Naming the optimal blowing trajectory according to the product type, fixture information, and machine information to obtain the named optimal blowing trajectory;

[0049] The named optimal blowing trajectory is stored in the database.

[0050] From the above description, it can be seen that the optimal blowing trajectory is named according to the product type, fixture information and machine information, so as to quickly determine the appropriate blowing trajectory according to the product type, fixture information and machine information.

[0051] Please refer to Figure 2 , a deep learning-based intelligent cleaning terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0052] Determine the product type, chip status, fixture information, and machine information of the test product;

[0053] Using deep learning technology to perform debris cleaning simulation on the test product according to the product type, debris state, fixture information, and machine information, obtain the corresponding optimal blowing trajectory, and store the optimal blowing trajectory in a database;

[0054] Selecting a target blowing trajectory from the database according to the product type, debris status, fixture information, and machine information of the product to be cleaned;

[0055] The debris on the product to be cleaned is cleaned according to the target blowing trajectory to obtain a cleaned product.

[0056] From the above description, it can be seen that the beneficial effects of the present invention are: using deep learning technology to simulate the debris cleaning of the test product according to the product type, debris state, fixture information and machine information, and obtain the corresponding optimal blowing trajectory, and store the optimal blowing trajectory in the database, and select the target blowing trajectory from the database according to the product type, debris state, fixture information and machine information of the product to be cleaned, and clean the debris on the product to be cleaned according to the target blowing trajectory to obtain the cleaned product. It is no longer like the prior art that uses manual blowing cleaning, but uses deep learning technology to learn different blowing trajectories according to different product types, debris states, fixture information and machine information. Subsequently, the blowing trajectory can be selected according to the information corresponding to the product to be cleaned for targeted cleaning, thereby efficiently cleaning the residual debris on the product.

[0057] Furthermore, the use of deep learning technology to perform debris cleaning simulation on the test product according to the product type, debris state, fixture information and machine information to obtain the corresponding optimal blowing trajectory includes:

[0058] Acquire a status image of the test product, wherein the status image includes a debris image;

[0059] Acquire a simulated blowing trajectory corresponding to the state image until the debris image does not exist in the state image;

[0060] Combining all the simulated blowing trajectories into an initial blowing trajectory;

[0061] Performing deep learning on the initial blowing trajectory according to the product type, debris state, fixture information, and machine information to obtain a learned blowing trajectory;

[0062] Cleaning debris from a preset number of inspection products according to the learned blowing trajectory to obtain a corresponding cleaning accuracy and cleaning time;

[0063] Determine whether the cleaning accuracy reaches a first preset value and whether the cleaning time is less than a second preset value. If so, determine the learned blowing trajectory as the optimal blowing trajectory. If not, return to execute the step of obtaining the status image of the test product.

[0064] From the above description, it can be seen that the debris state of each time and its corresponding simulated blowing trajectory as well as the debris state obtained after the simulated blowing trajectory is run are used as the learning set. When deep learning of the blowing trajectory is performed, the features in the simulated blowing trajectory are matched with the debris state according to the cleaning accuracy and cleaning time, and the optimal blowing trajectory is determined, which can accurately clean the debris on the product, thereby improving the cleaning quality of residual debris.

[0065] Furthermore, the determining of the product type, debris status, fixture information and machine information of the test product includes:

[0066] Collect a large number of different debris states to form a debris state library;

[0067] The selecting a target blowing trajectory from the database according to the product type, debris status, fixture information, and machine information of the product to be cleaned includes:

[0068] Determine the debris state of the product to be cleaned using visual inspection technology based on the debris state library;

[0069] Determine the product type, fixture information, and machine information of the product to be cleaned;

[0070] The target blowing trajectory is selected from the database according to the product type, debris status, fixture information and machine information of the product to be cleaned.

[0071] From the above description, it can be seen that by collecting a large number of different debris states to form a debris state library, the debris state of the product to be cleaned can be accurately determined by using visual inspection technology based on the debris state library, and the corresponding blowing trajectory can be used for cleaning according to different product types, debris states, fixture information and machine information, thereby improving the intelligence level of product debris cleaning.

[0072] Further, the cleaning of debris on the product to be cleaned according to the target blowing trajectory to obtain the cleaned product includes:

[0073] Cleaning the debris on the product to be cleaned according to the target blowing trajectory to obtain an initially cleaned product;

[0074] Using visual inspection technology to detect whether there are debris in the product after the initial cleaning, if not, it is determined that the cleaning is complete and the cleaned product is obtained, and if there are debris, the position corresponding to the debris that has not been cleaned is determined;

[0075] The product after the initial cleaning is cleaned by fixed-point directional blowing according to the positions corresponding to the debris that has not been cleaned up, until there is no debris in the product after the initial cleaning, so as to obtain a cleaned product.

[0076] From the above description, it can be seen that after obtaining the product after initial cleaning, visual inspection technology is used to detect whether there is any debris on the product after initial cleaning. If so, the position corresponding to the debris that has not been cleaned up is determined, and cleaning is performed again until there is no debris, thereby ensuring that the debris on the product is completely cleaned up.

[0077] Furthermore, storing the optimal blowing trajectory in a database includes:

[0078] Naming the optimal blowing trajectory according to the product type, fixture information, and machine information to obtain the named optimal blowing trajectory;

[0079] The named optimal blowing trajectory is stored in the database.

[0080] From the above description, it can be seen that the optimal blowing trajectory is named according to the product type, fixture information and machine information, so as to quickly determine the appropriate blowing trajectory according to the product type, fixture information and machine information.

[0081] The above-mentioned intelligent cleaning method based on deep learning of the present invention can be applied to a machining workshop, and is described below through a specific implementation method:

[0082] Embodiment 1

[0083] Please refer to Figure 1 and Figure 3 , an intelligent cleaning method based on deep learning in this embodiment comprises the steps of:

[0084] S1. Collect a large number of different debris states to form a debris state library;

[0085] Specifically, a large number of different debris states are collected using visual inspection technology to form a debris state library;

[0086] S2. Determine the product type, debris status, fixture information, and machine information of the test product;

[0087] In an optional embodiment, the product types include notebook metal shells, mobile phone metal shells and watch metal shells, the debris states include flaky aluminum chips, fine needle-shaped aluminum chips and rolled aluminum chips, the jig information includes 1-clip jigs, 2-clip jigs, 3-clip jigs, 4-clip jigs, 5-clip jigs, 6-clip jigs and 7-clip jigs, and the machine information includes machine brands;

[0088] S3, using deep learning technology to perform debris cleaning simulation on the test product according to the product type, debris state, fixture information and machine information, obtain the corresponding optimal blowing trajectory, and store the optimal blowing trajectory in the database, such as Figure 3 As shown, specifically including:

[0089] S31, acquiring a state image of the test product, wherein the state image includes a debris image;

[0090] S32, obtaining a simulated blowing trajectory corresponding to the state image; specifically, the simulated blowing trajectory is a trajectory recorded by a person operating the robot arm movement;

[0091] S33, repeatedly executing S31-S32 until the debris image does not exist in the state image;

[0092] Specifically, the operator holds the robot arm and runs along the default Z-shaped air blowing trajectory, and the robot arm records this path, so as to obtain a simulated air blowing trajectory corresponding to the state image, until the debris image does not exist in the state image;

[0093] Wherein, visual detection technology may be used to determine whether the debris image exists in the state image;

[0094] For example, if the product type of the tested product is a metal shell for a notebook, the debris state is flaky debris, the fixture state is a 3-clamp fixture, and the machine information is a group brand, the operator holds the robot arm and runs it along the default Z-shaped air blowing trajectory, and the robot arm records this path. When the vision detects that there is a debris image in the state image, the operator holds the robot arm and runs it again along a different air blowing trajectory, and the robot arm records it again, until the vision detects that there is no debris image in the state image;

[0095] S34, combining all the simulated blowing trajectories into an initial blowing trajectory;

[0096] S35, performing deep learning on the initial blowing trajectory according to the product type, debris state, fixture information, and machine information to obtain a learned blowing trajectory;

[0097] In an optional embodiment, the initial blowing trajectory has a corresponding relationship with the debris state, that is, each of the simulated blowing trajectories in the initial blowing trajectory corresponds to two debris states, one is the debris state before the simulated blowing trajectory is executed, and the other is the debris state after the simulated blowing trajectory is executed, so that the deep learning algorithm can learn the correspondence between the debris state and the specific characteristics of the simulated blowing trajectory through the characteristics of the changes of the simulated blowing trajectories adjacent in time and the different states of the debris;

[0098] In an optional implementation, the deep learning algorithm is pytorch yolov5.

[0099] S36, performing debris cleaning on a preset number of inspection products according to the learned blowing trajectory, and obtaining a corresponding cleaning accuracy and cleaning time;

[0100] The preset number can be set according to actual conditions. In this embodiment, the preset number is 100 units.

[0101] S37, judging whether the cleaning accuracy reaches a first preset value and whether the cleaning time is less than a second preset value, if so, determining the learned blowing trajectory as the optimal blowing trajectory, if not, returning to execute the step of obtaining the state image of the test product;

[0102] Wherein, the first preset value is 100%, and the second preset value is 15 seconds;

[0103] S38, storing the optimal blowing trajectory in a database, specifically comprising:

[0104] S381, naming the optimal blowing trajectory according to the product type, fixture information and machine information, and obtaining the named optimal blowing trajectory;

[0105] For example, if the product type is a notebook metal shell, the fixture information is a 3-clip fixture, and the machine information is a FANUC brand, then the corresponding optimal blowing trajectory can be named mobile phone metal shell 4-clip-2 trajectory;

[0106] S382, storing the named optimal blowing trajectory in a database;

[0107] S4, selecting a target blowing trajectory from the database according to the product type, debris status, fixture information, and machine information of the product to be cleaned, specifically including:

[0108] S41, determining the debris state of the product to be cleaned using visual detection technology based on the debris state library;

[0109] S42, determining the product type, fixture information, and machine information of the product to be cleaned;

[0110] S43, selecting a target blowing trajectory from the database according to the product type, debris status, fixture information, and machine information of the product to be cleaned;

[0111] S5, cleaning the debris on the product to be cleaned according to the target blowing trajectory to obtain a cleaned product, specifically comprising:

[0112] S51, cleaning the debris on the product to be cleaned according to the target blowing trajectory to obtain an initially cleaned product;

[0113] S52, using visual inspection technology to detect whether there are debris in the product after the initial cleaning, if not, determining that the cleaning is complete and obtaining the cleaned product, if yes, determining the position corresponding to the debris that has not been cleaned up;

[0114] S53, performing fixed-point directional blowing cleaning on the initially cleaned product according to the positions corresponding to the uncleaned debris, until no debris exists in the initially cleaned product, thereby obtaining a cleaned product.

[0115] Embodiment 2

[0116] Please refer to Figure 2 In this embodiment, a deep learning-based intelligent cleaning terminal includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, each step of the deep learning-based intelligent cleaning method in the first embodiment is implemented.

[0117] In summary, the present invention provides an intelligent cleaning method and terminal based on deep learning, which determine the product type, debris state, fixture information and machine information of a test product; use deep learning technology to perform debris cleaning simulation on the test product according to the product type, debris state, fixture information and machine information to obtain the corresponding optimal blowing trajectory, and store the optimal blowing trajectory in a database; select a target blowing trajectory from the database according to the product type, debris state, fixture information and machine information of the product to be cleaned; clean the debris on the product to be cleaned according to the target blowing trajectory to obtain a cleaned product, and when performing deep learning of the blowing trajectory, the optimal blowing trajectory will be determined according to the cleaning accuracy and cleaning time, which can accurately clean the debris on the product and improve the cleaning quality of the residual debris. The blowing trajectory can be selected according to the information corresponding to the product to be cleaned for targeted cleaning, thereby efficiently cleaning the residual debris on the product.

[0118] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's specification and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An intelligent cleaning method based on deep learning, It is characterized in that Includes steps: Determine the product type, chip status, fixture information, and machine information of the test product; Using deep learning technology to perform debris cleaning simulation on the test product according to the product type, debris state, fixture information, and machine information, obtain the corresponding optimal blowing trajectory, and store the optimal blowing trajectory in a database; Selecting a target blowing trajectory from the database according to the product type, debris status, fixture information, and machine information of the product to be cleaned; Cleaning the debris on the product to be cleaned according to the target blowing trajectory to obtain a cleaned product; The use of deep learning technology to perform debris cleaning simulation on the test product according to the product type, debris state, fixture information and machine information to obtain the corresponding optimal blowing trajectory includes: Acquire a status image of the test product, wherein the status image includes a debris image; Acquire a simulated blowing trajectory corresponding to the state image until the debris image does not exist in the state image; Combining all the simulated blowing trajectories into an initial blowing trajectory; Performing deep learning on the initial blowing trajectory according to the product type, debris state, fixture information, and machine information to obtain a learned blowing trajectory; Cleaning debris from a preset number of inspection products according to the learned blowing trajectory to obtain a corresponding cleaning accuracy and cleaning time; Determine whether the cleaning accuracy reaches a first preset value and whether the cleaning time is less than a second preset value, if so, determine the learned blowing trajectory as the optimal blowing trajectory, if not, return to execute the step of obtaining the state image of the test product; The acquiring of the simulated blowing trajectory corresponding to the state image until the debris image does not exist in the state image comprises: After the operator holds the robot arm and runs along the default Z-shaped air blowing trajectory once, the robot arm records this path, thereby obtaining a simulated air blowing trajectory corresponding to the state image, until the debris image does not exist in the state image; Each of the simulated air blowing trajectories in the initial air blowing trajectory corresponds to two of the debris states, and the two debris states include a debris state before the simulated air blowing trajectory is executed and a debris state after the simulated air blowing trajectory is executed; The initial blowing trajectory is deeply learned according to the product type, debris state, fixture information and machine information, and the learned blowing trajectory includes: Based on the product type, fixture information and machine information, the corresponding learning of the debris state and the specific characteristics of the simulated blowing trajectory is carried out through the characteristics of the changes in the simulated blowing trajectory adjacent in time and the different states of the debris, and the learned blowing trajectory is obtained.

2. According to the intelligent cleaning method based on deep learning according to claim 1, It is characterized in that The determining of the product type, debris status, fixture information and machine information of the test product includes: Collect a large number of different debris states to form a debris state library; The selecting a target blowing trajectory from the database according to the product type, debris status, fixture information, and machine information of the product to be cleaned includes: Determine the debris state of the product to be cleaned using visual inspection technology based on the debris state library; Determine the product type, fixture information, and machine information of the product to be cleaned; The target blowing trajectory is selected from the database according to the product type, debris status, fixture information and machine information of the product to be cleaned.

3. According to the intelligent cleaning method based on deep learning according to claim 1, It is characterized in that The cleaning of debris on the product to be cleaned according to the target blowing trajectory to obtain the cleaned product comprises: Cleaning the debris on the product to be cleaned according to the target blowing trajectory to obtain an initially cleaned product; Using visual inspection technology to detect whether there are debris in the product after the initial cleaning, if not, it is determined that the cleaning is complete and the cleaned product is obtained, and if there are debris, the position corresponding to the debris that has not been cleaned is determined; The product after the initial cleaning is cleaned by fixed-point directional blowing according to the positions corresponding to the debris that has not been cleaned up, until there is no debris in the product after the initial cleaning, so as to obtain a cleaned product.

4. According to claim 1, an intelligent cleaning method based on deep learning, It is characterized in that The storing of the optimal blowing trajectory in a database comprises: Naming the optimal blowing trajectory according to the product type, fixture information, and machine information to obtain the named optimal blowing trajectory; The named optimal blowing trajectory is stored in the database.

5. A deep learning-based intelligent cleaning terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, the following steps are implemented: Determine the product type, chip status, fixture information, and machine information of the test product; Using deep learning technology to perform debris cleaning simulation on the test product according to the product type, debris state, fixture information, and machine information, obtain the corresponding optimal blowing trajectory, and store the optimal blowing trajectory in a database; Selecting a target blowing trajectory from the database according to the product type, debris status, fixture information, and machine information of the product to be cleaned; Cleaning the debris on the product to be cleaned according to the target blowing trajectory to obtain a cleaned product; The use of deep learning technology to perform debris cleaning simulation on the test product according to the product type, debris state, fixture information and machine information to obtain the corresponding optimal blowing trajectory includes: Acquire a status image of the test product, wherein the status image includes a debris image; Acquire a simulated blowing trajectory corresponding to the state image until the debris image does not exist in the state image; Combining all the simulated blowing trajectories into an initial blowing trajectory; Performing deep learning on the initial blowing trajectory according to the product type, debris state, fixture information, and machine information to obtain a learned blowing trajectory; Cleaning debris from a preset number of inspection products according to the learned blowing trajectory to obtain a corresponding cleaning accuracy and cleaning time; Determine whether the cleaning accuracy reaches a first preset value and whether the cleaning time is less than a second preset value, if so, determine the learned blowing trajectory as the optimal blowing trajectory, if not, return to execute the step of obtaining the state image of the test product; The acquiring of the simulated blowing trajectory corresponding to the state image until the debris image does not exist in the state image comprises: After the operator holds the robot arm and runs along the default Z-shaped air blowing trajectory once, the robot arm records this path, thereby obtaining a simulated air blowing trajectory corresponding to the state image, until the debris image does not exist in the state image; Each of the simulated air blowing trajectories in the initial air blowing trajectory corresponds to two of the debris states, and the two debris states include a debris state before the simulated air blowing trajectory is executed and a debris state after the simulated air blowing trajectory is executed; The initial blowing trajectory is deeply learned according to the product type, debris state, fixture information and machine information, and the learned blowing trajectory includes: Based on the product type, fixture information and machine information, the corresponding learning of the debris state and the specific characteristics of the simulated blowing trajectory is carried out through the characteristics of the changes in the simulated blowing trajectory adjacent in time and the different states of the debris, and the learned blowing trajectory is obtained.

6. According to claim 5, a smart cleaning terminal based on deep learning, It is characterized in that The determining of the product type, debris status, fixture information and machine information of the test product includes: Collect a large number of different debris states to form a debris state library; The selecting a target blowing trajectory from the database according to the product type, debris status, fixture information, and machine information of the product to be cleaned includes: Determine the debris state of the product to be cleaned using visual inspection technology based on the debris state library; Determine the product type, fixture information, and machine information of the product to be cleaned; The target blowing trajectory is selected from the database according to the product type, debris status, fixture information and machine information of the product to be cleaned.

7. The deep learning-based intelligent cleaning terminal according to claim 5, It is characterized in that The cleaning of debris on the product to be cleaned according to the target blowing trajectory to obtain the cleaned product comprises: Cleaning the debris on the product to be cleaned according to the target blowing trajectory to obtain an initially cleaned product; Using visual inspection technology to detect whether there are debris in the product after the initial cleaning, if not, it is determined that the cleaning is complete and the cleaned product is obtained, and if there are debris, the position corresponding to the debris that has not been cleaned is determined; The product after the initial cleaning is cleaned by fixed-point directional blowing according to the positions corresponding to the debris that has not been cleaned up, until there is no debris in the product after the initial cleaning, so as to obtain a cleaned product.

8. The deep learning-based intelligent cleaning terminal according to claim 5, It is characterized in that The storing of the optimal blowing trajectory in a database comprises: Naming the optimal blowing trajectory according to the product type, fixture information, and machine information to obtain the named optimal blowing trajectory; The named optimal blowing trajectory is stored in the database.

Images