System and methods for upsampling of decompressed data after lossy compression using a neural network
A system and method for complex-valued radar image compression integrates AI-based techniques to enhance compression quality. It incorporates a novel AI deblocking network composed of convolutional layers for feature extraction and a channel-wise transformer with attention to capture complex inter-channel dependencies. The convolutional layers extract multi-dimensional features from the complex-valued radar image, while the channel-wise transformer learns global inter-channel relationships. This hybrid approach addresses both local and global features, mitigating compression artifacts and improving image quality. The model's outputs enable effective complex-valued radar image reconstruction, achieving advanced compression while preserving crucial information for accurate analysis.
1 . A system for N-channel data compression with predictive recovery, comprising:
a computing device comprising at least a memory, a processor, and a network interface;
a trained neural network configured to recover inter-channel relationships from N-channel data, where N≥2, the N channels exhibiting a correlation relationship;
an encoder comprising a first plurality of programming instructions stored in the memory and operable on the processor, wherein the first plurality of programming instructions cause the computing device to:
receive N-channel input data; and
perform lossy compression on the N-channel input data to generate a compressed bitstream; and
a decoder comprising a second plurality of programming instructions stored in the memory and operable on the processor, wherein the second plurality of programming instructions cause the computing device to:
receive the compressed bitstream;
decompress the compressed bitstream to predict the N channels; and
process the predicted N channels through the trained neural network to recover information lost during lossy compression by leveraging the correlation relationship between the N channels.
2 . The system of claim 1 , wherein the N-channel input data comprises image data with N≥3 color channels.
3 . The system of claim 2 , wherein the N channels comprise red, green, and blue color channels.
4 . The system of claim 1 , wherein the N-channel input data comprises sensor network data from N≥2 sensors exhibiting temporal correlation.
5 . The system of claim 4 , wherein the sensors comprise at least two of: pressure sensors, thermal sensors, barometers, wind speed sensors, humidity sensors, and air quality sensors.
6 . The system of claim 1 , wherein the correlation relationship comprises spatial correlation between the N channels.
7 . The system of claim 1 , wherein the correlation relationship comprises temporal correlation between the N channels.
8 . The system of claim 1 , wherein the trained neural network comprises:
N parallel processing branches, each branch configured to process one of the N channels;
a plurality of convolutional layers in each branch; and
an N-channel wise transformer configured to combine outputs from the N branches.
9 . The system of claim 8 , wherein the N-channel wise transformer comprises a self-attention mechanism configured to capture dependencies between the N channels.
10 . The system of claim 8 , wherein each branch further comprises a plurality of residual blocks.
11 . The system of claim 1 , wherein the trained neural network is trained using a two-stage process, each stage utilizing a specific loss function.
12 . The system of claim 11 , wherein at least one loss function comprises mean squared error applied to the N channels.
13 . The system of claim 1 , wherein the encoder further comprises a quantizer configured to perform uniform quantization on the N channels prior to lossy compression.
14 . The system of claim 1 , wherein the decoder further comprises a dequantizer configured to restore the N channels to an initial dynamic range after processing by the trained neural network.
15 . The system of claim 1 , wherein the lossy compression comprises High Efficiency Video Coding applied independently to each of the N channels.
16 . The system of claim 1 , wherein the encoder further comprises a lossless compactor configured to apply lossless data compaction to the compressed bitstream.
17 . The system of claim 16 , wherein the decoder further comprises a lossless decompactor configured to restore the compressed bitstream from a compacted format prior to decompression.
18 . The system of claim 1 , wherein the trained neural network comprises deformable convolutional layers configured to adaptively sample input features based on learned offsets.
19 . The system of claim 1 , wherein processing the predicted N channels through the trained neural network comprises removing compression artifacts introduced during the lossy compression.
20 . The system of claim 1 , wherein the trained neural network comprises a multi-scale attention block configured to capture inter-channel dependencies at multiple spatial scales.