Vehicle neural network

1. A computer, comprising a processor; and

a memory, the memory including instructions to be executed by the processor to:

generate a first color image of a road environment;

determine one or more singular value decompositions of one or more of a red, green, and blue channel of the first color image;

obtain one or more modified singular value decompositions by modifying respective ones of the singular value decompositions by a non-linear equation corresponding to modified lighting conditions;

reconstruct a second color image based on the modified one or more singular value decompositions;

train a deep neural network based on the second color image; and

operate a vehicle based on the deep neural network.

2. The computer of claim 1 , the instructions including further instructions to generate the first color image by one or more of acquiring vehicle sensor data and generating synthetic image data.

3. The computer of claim 1 , wherein the modified lighting conditions correspond to darker lighting conditions including overcast lighting, evening or night.

4. The computer of claim 1 , the instructions including further instructions to determine the one or more singular value decompositions including with a matrix equation X=UΣV T , wherein X is an array that includes one of a red, green, or blue color channel of the first color image, U and V are left and right eigenvectors of array X, and Σ is a diagonal matrix of singular values.

5. The computer of claim 1 , the instructions including further instructions to modify the one or more singular value decompositions based on a non-linear equation S={s−n for s≥S, else 0}, wherein s is a singular value, n is an integer and S is a set of singular values.

6. The computer of claim 1 , wherein the deep neural network is convolutional neural network.

7. The computer of claim 1 , the instructions including further instructions to train the deep neural network to identify objects in the road environment including a roadway, another vehicle, and a pedestrian.

8. The computer of claim 1 , the instructions including further instructions to operate the vehicle based on the deep neural network by inputting vehicle sensor data to the deep neural network which outputs data regarding objects in the road environment.

9. The computer of claim 8 , the instructions including further instructions to operate the vehicle based on determining a vehicle path based on the data regarding objects in the road environment.

10. The computer of claim 9 , the instructions including further instructions to operate the vehicle along the vehicle path by controlling vehicle powertrain, steering and brakes.

11. A method, comprising:

generating a first color image of a road environment;

determining one or more singular value decompositions of one or more of a red, green, and blue channel of the first color image;

obtaining one or more of modified singular value decompositions by modifying respective ones of the singular value decompositions by a non-linear equation corresponding to modified lighting conditions;

reconstructing a second color image based on the modified one or more singular value decompositions;

training a deep neural network based on the second color image; and

operating a vehicle based on the deep neural network.

12. The method of claim 11 , further comprising generating the first color image by one or more of acquiring vehicle sensor data and generating synthetic image data.

13. The method of claim 11 , wherein the modified lighting conditions correspond to darker lighting conditions including overcast lighting, evening or night.

14. The method of claim 11 , further comprising determining one or more singular value decompositions including with a matrix equation X=UΣV T , wherein X is an array that includes one of a red, green, or blue color channel of the first color image, U and V are left and right eigenvectors of an array X, and Σ is a diagonal matrix of singular values.

15. The method of claim 11 , further comprising modifying the one or more singular value decompositions based on a non-linear equation S={s−n for s≥S, else 0}, wherein s is a singular value, n is an integer and S is a set of singular values.

16. The method of claim 11 , wherein the deep neural network is convolutional neural network.

17. The method of claim 11 , further comprising training the deep neural network to identify objects in the road environment including a roadway, another vehicle, and a pedestrian.

18. The method of claim 11 , further comprising operating the vehicle based on the deep neural network by inputting vehicle sensor data to the deep neural network which outputs object data regarding objects in the road environment.

19. The method of claim 18 , further comprising determining a vehicle path based on the object data regarding objects in the road environment.

20. The method of claim 19 , further comprising operating the vehicle along the vehicle path by controlling vehicle powertrain, steering and brakes.