Neural vending machine

1. A method of dispensing a product (n) to a user comprising:

displaying an initial stock (S) of several products (n) on a display shelf ( 34 ) in a locked self-service dispensing machine ( 28 );

authorizing access to said user for selecting and removing any one of said products (n) displayed in said machine ( 28 );

using a camera ( 42 ) mounted inside said machine ( 28 ) with horizontal and vertical field-of-view to capture images ( 50 , 52 ) of the entire stock (S) of products (n) displayed on said shelf ( 34 );

identifying any product (n) removed from said display shelf ( 34 ) by said user not by detecting said removed product ( 13 ) itself, but by comparing images ( 50 , 52 ) of said displayed stock (S) captured by said camera before and after said product is removed by said user to determine any product missing in said image ( 52 ) of the remaining stock; and

accounting said missing product ( 13 ) to said user as being removed by said user from said machine ( 28 ).

2. A method of dispensing a product (n) to a user comprising:

displaying an initial stock (S) of several products (n) on a display shelf ( 34 ) in a locked self-service dispensing machine ( 28 );

authorizing access to said user for selecting and removing any one of said products (n) displayed in said machine ( 28 );

imaging ( 50 , 52 ) said displayed stock (S) before and after said product ( 13 ) is removed therefrom by said user;

identifying from said imaging ( 50 , 52 ) all products (n) in both said initial stock (S) before product removal and in said remaining stock (S-(S-P)) after product removal;

identifying any product (n) removed from said display shelf ( 34 ) by said user not by detecting said removed product ( 13 ) itself, but by comparing said products (n) identified in said initial and remaining stocks to determine any missing product ( 13 ) therebetween, and thereby designate said missing product ( 13 ) as said removed product ( 13 ); and

accounting said missing product ( 13 ) to said user as being removed by said user from said machine ( 28 ).

3. A method according to claim 2 further comprising:

pre-imaging ( 50 ) said initial stock (S) of products (n) before product removal;

identifying said initial stock (S) of products (n) from said pre-image ( 50 );

post-imaging ( 52 ) said remaining stock (S-(S-P)) of products (n) after product removal;

identifying said remaining stock (S-(S-P)) of products (ii) from said post-image ( 52 ); and

comparing said identified remaining stock (S-(S-P)) and said identified initial stock (S) to identify said missing product ( 13 ).

4. A method according to claim 3 further comprising:

deploying an Artificial Neural Network (ANN) trained to both detect and recognize each product (n) in said stock pre-image ( 50 );

deploying an Artificial Neural Network (ANN) trained to both detect and recognize each product (n) in said stock post-image ( 52 ); and

comparing said ANN-recognized stock products (n) between said post-image ( 52 ) and said pre-image ( 50 ) to identify said missing product ( 13 ).

5. A method according to claim 4 wherein said ANNs are pre-trained prior to use in dispensing said products (n), with said pre-training comprising:

imaging an inventory (N) of a multitude of products (n) including said initial product stock (S); and

training said ANNs to detect and recognize from said imaging each product (n) in said inventory (N) based on correspondingly developed neural signatures (X(n)).

6. A method according to claim 4 wherein the same pretrained ANN is used to detect and recognize said products (n) in both said pre-image ( 50 ) and post-image ( 52 ).

7. A method according to claim 4 wherein:

two different ANNs (−1,−2) are deployed in parallel in syndicate pooling evaluation to independently detect and recognize said products (n) in both said pre-image ( 50 ) and post-image ( 52 ); and

product recognition must agree for both different ANNs (−1,−2) for both said stock pre-image ( 50 ) and stock post-image ( 52 ) to identify said missing product ( 13 ).

8. A method according to claim 7 wherein said two different ANNs comprise:

a Single Shot Detector (SSD-ANN-1); and

a Region-based Convolutional Neural Network (RCNN-ANN-2).

9. A method according to claim 4 further comprising:

imaging an inventory (N) of a multitude of products (n) including said initial product stock (S);

creating a secondary signature (Y(n)) for each product (n) in said inventory based on product appearance;

deploying a Secondary Visual Recognition System (SVRS 58 ) to identify from said secondary signature (Y(n)) each product (n) in both said stock pre-image ( 50 ) and in said stock post-image ( 52 ); and

comparing said stock pre-image ( 50 ) and stock post-image ( 52 ) to identify said missing product ( 13 ) based on said secondary signature (Y(n)) thereof.

10. A method according to claim 9 wherein said secondary signature (Y(n)) is a color signature of said products (n), and said SVRS ( 58 ) includes Binary Large Object (BLOB) detection of said color signatures (Y(n)).

11. A method according to claim 9 wherein said secondary signature (Y(n)) is text printed on said products (n), and said SVRS ( 58 ) includes Optical Character Recognition (OCR) thereof.

12. A method according to claim 4 further comprising:

displaying said initial stock (S) of products (n) on said display shelf ( 34 ) a locked display cabinet ( 30 ) inside a vending machine ( 28 );

authorizing access to said user for purchasing from said display cabinet ( 30 );

pre-image ( 50 ) and identify therefrom said initial stock (S) of products (n) before unlocking said cabinet ( 30 );

unlocking said cabinet ( 30 ) to allow access thereto by said user for removing any one or more of said displayed products (n);

post-image ( 52 ) and identify therefrom said remaining stock (S-(S-P)) of products (n) after a product ( 13 ) has been removed by said user;

matching products (n) identified in said pre-image ( 50 ) and said post-image ( 52 ) to determine said product ( 13 ) missing from said post-image ( 52 ); and

charging payment to said user for said missing product ( 13 ).

13. A method according to claim 12 further comprising:

displaying said products (n) on multiple shelves ( 34 ) inside said display cabinet ( 30 ) behind a locked display door ( 32 );

mounting a digital camera ( 42 ) inside said cabinet ( 30 ) with horizontal and vertical field-of-view to capture images ( 50 , 52 ) of the entire stock (S) of products (n) displayed on said multiple shelves ( 34 );

joining said camera ( 42 ) to a digital computer ( 44 ) housed inside said vending machine, with said computer ( 44 ) including said trained ANN programmed therein;

pre-image ( 50 ) using said camera ( 42 ) and identify using said trained ANN said initial stock (S) of products (n) displayed on said multiple shelves ( 34 ) before said user opens said door ( 32 );

post-image ( 52 ) using said camera ( 42 ) and identify using said trained ANN said remaining stock (S-(S-P)) of products (n) displayed on said multiple shelves ( 34 ) after said user removes a product ( 13 ) and closes said door ( 32 ); and

determine said product ( 13 ) missing from said post-image ( 52 ) and charge payment therefor to said user.

14. A method according to claim 13 wherein the same pretrained ANN is used to detect and recognize said products (n) in both said pre-image ( 50 ) and post-image ( 52 ).

15. A method according to claim 13 wherein:

two different ANNs (ANN-1, ANN-2) are deployed in parallel n syndicate pooling evaluation to independently detect and recognize said products (n) in both said pre-image ( 50 ) and post-image ( 52 ); and

product recognition must agree for both different ANNs for both said stock pre-image ( 50 ) and stock post-image ( 52 ) to identify said missing product ( 13 ).

16. A method according to claim 15 wherein said two different ANNs comprise:

a Single Shot Detector (SSD-ANN-1); and

a Region-based Convolutional Neural Network (RCNN-ANN-2).

17. A method according to claim 16 further comprising:

imaging an inventory (N) of a multitude of products (n) including said initial product stock (S);

creating a secondary signature (Y(n)) for each product (n) in said inventory based on product appearance;

deploying a Secondary Visual Recognition System (SVRS 58 ) to identify from said secondary signature (Y(n) each product (n) in both said stock pre-image ( 50 ) and in said stock post-image ( 52 ); and

comparing said stock pre-image ( 50 ) and stock post-image ( 52 ) to identify said missing product ( 13 ) based on said secondary signature (Y(n)) thereof.

18. A method according to claim 13 further comprising:

mounting said vending machine ( 28 ) to an aircraft fuselage ( 22 ) inside a passenger cabin ( 26 ), with said display cabinet ( 30 ) accessible to passengers during flight; and

said vending machine ( 28 ) having minimal complexity and weight as characterized by the express absence of systems for directly identifying and automatically dispensing any product from said display cabinet including barcode readers ( 62 ), Radio-Frequency Identification (RFD) detectors ( 64 ), and mechanically-driven dispensing chutes ( 66 ).

19. A method according to claim 1 further comprising:

displaying said initial stock (S) of products (n) in random locations on multiple display shelves ( 34 ) in a display cabinet ( 30 ) in an automated vending machine ( 28 ) having a locked display door ( 32 ) through which said products (n) are visible;

said cabinet ( 30 ) including a digital camera ( 42 ) having a field-of-view including the entire stock (S) of products (n) displayed on said shelves ( 34 );

said camera ( 42 ) being operatively joined to a digital computer ( 44 ) configured in software for identifying said product ( 13 ) removed by said user from said cabinet ( 30 ) by comparing pre and post images ( 50 , 52 ) taken by said camera ( 42 ) of said displayed stock (S) before and after said product removal to determine any product ( 13 ) missing in said post-image ( 52 ) of remaining stock (S-(S-P));

said computer ( 44 ) being further configured for authorizing user access to said locked cabinet ( 30 ), unlocking and re-locking said door ( 32 ) before and after product removal, and processing payment from said user for said missing product ( 13 ).

20. A method according to 19 wherein said computer software includes:

a first Artificial Neural Network (ANN-1) pre-trained to both detect and recognize each product (n) in said stock images ( 50 , 52 );

a second Artificial Neural Network (ANN-2) differently configured than said first ANN, and pre-trained to both detect and recognize each product (n) in said stock images ( 50 , 52 );

said first and second ANNs being joined in parallel in syndicate pooling evaluation to independently detect and recognize said products (n) in both said pre-image ( 50 ) and post-image ( 52 );

said software is further configured for comparing said ANN-recognized stock products (n) between said post-image ( 52 ) and said pre-image ( 50 ) to identify said missing product ( 13 ); and

said product recognition by said first and second ANNs must agree for both said stock pre-image ( 50 ) and said stock post-image ( 52 ) to identify said missing product ( 13 ).

21. An automated vending machine ( 28 ) comprising:

a display cabinet ( 30 ) having a locked display door ( 32 ), and including multiple display shelves ( 34 ) for displaying through said door ( 30 ) an initial stock (S) of several products (n) for sale;

a digital camera ( 42 ) mounted inside said cabinet ( 30 ) with a field-of-view including the entire stock (S) of products (n) displayed on said shelves ( 34 );

a digital computer ( 44 ) operatively joined to said camera ( 42 ), and configured in software for identifying any product ( 13 ) removed from said cabinet ( 30 ) by a user not by detecting said removed product ( 13 ) itself, but by comparing pre and post images ( 50 , 52 ) taken by said camera ( 42 ) of said displayed stock (S) before and after said product removal to determine any product ( 13 ) missing in said post-image ( 52 ) of remaining stock (S-(S-P); and

said computer ( 44 ) further configured to authorize access to said user, unlock and re-lock said door ( 32 ) before and after product removal, and processing payment from said user for said missing product ( 13 ).

22. A vending machine ( 28 ) according to claim 21 wherein said computer software includes:

a first Artificial Neural Network (ANN-1) pre-trained to both detect acid recognize each product (n) in said stock images ( 50 , 52 );

a second Artificial Neural Network (ANN-2) differently configured than said first ANN, and pre-trained to both detect and recognize each product (n) in said stock images ( 50 , 52 );

said first and second ANNs being joined in parallel in syndicate pooling evaluation to independently detect and recognize said products (n) in both said pre-image ( 50 ) and post-image ( 52 );

said software is further configured for comparing said ANN-recognized stock products (n) between said post-image ( 52 ) and said pre-image ( 50 ) to identify said missing product ( 13 ); and

said product recognition by said first and second ANNs must agree for both said stock pre-image ( 50 ) and said stock post-image ( 52 ) to identify said missing product ( 13 ).