Agricultural pattern analysis system
A pattern recognition system including an image gathering unit that gathers at least one digital representation of a field, an image analysis unit that pre-processes the at least one digital representation of a field, an annotation unit that provides a visualization of at least one channel for each of the at least one digital representation of the field, where the image analysis unit generates a plurality of image samples from each of the at least one digital representation of the field, and the image analysis unit splits each of the image samples into a plurality of categories.
1. A pattern recognition system including:
an image gathering unit that gathers digital at least one representation of a field;
an image analysis unit that pre-processes the at least one representation of a field;
an annotation unit that provides a visualization of at least one channel for each of the at least one digital representation of the field;
wherein,
the each of the samples is 512 pixels by 512 pixels and each of the samples is randomly split using a train, val, test ratio,
the image analysis unit generates a plurality of image samples from each of the at least one digital representation of the field, and
the image analysis unit splits each of the image samples into a plurality of categories.
2. The pattern recognition system of claim 1 , wherein the annotation unit separates each digital representation into RGB, NIR and NDVI channels.
3. The pattern recognition system of claim 1 , wherein adjacent samples are compared for overlap.
4. The pattern recognition system of claim 3 , wherein a sample is discarded if the sample has more than 30% overlap with an adjacent sample.
5. The pattern recognition system of claim 1 , wherein the image analysis unit randomly splits each sample into at least three categories.
6. The pattern recognition system of claim 1 , wherein the image analysis unit generates a semantic map by applying a modified FPN model to each image sample.
7. The pattern recognition system of claim 6 , wherein the FPN model encoder is a ResNet.
8. The pattern recognition system of claim 7 , wherein the FPN decoder includes two 3×3 and one 1×1 convolution layer.
9. The pattern recognition system of claim 8 , wherein each 3×3 convolution layer includes batch normalization layer and a leaky ReLU layer.
10. A method of recognizing a pattern in an image by an image recognition unit including a processor and a memory, with a program in the memory executing the steps of:
gathering at least one digital representation of a field;
pre-processing the at least one representation of a field;
providing a visualization of at least one channel for each of the at least one digital representation of the field;
generating a plurality of image samples from each of the at least one digital representation of the field, and
splitting each of the image samples into a plurality of categories,
wherein,
each of the image samples is 512 pixels by 512 pixels and each of the samples is randomly split using a train, val, test ratio.
11. The method of claim 10 , including the step of separating each digital representation into RGB, NIR and NDVI channels.
12. The method of claim 10 , wherein adjacent samples are compared for overlap.
13. The method of claim 12 , wherein a sample is discarded if the sample has more than 30% overlap with an adjacent sample.
14. The method of claim 10 , including the step of randomly splitting each sample into at least three categories.
15. The method of claim 10 , including the step of generating a semantic map by applying a modified FPN model to each image sample.
16. The method of claim 15 , wherein the FPN model encoder is a ResNet.
17. The method of claim 16 , wherein the FPN decoder includes two 3×3 and one 1×1 convolution layer.
18. The method of claim 17 , wherein each 3×3 convolution layer includes batch normalization layer and a leaky ReLU layer.