Polarization intelligent sensing system and polarization intelligent sensing method

A polarization intelligent sensing system, comprising:

• a polarization optical imaging module, consisting of a polarization optical lens and a polarization detector, configured to perform polarization imaging on a target scene to obtain a polarization image;
• a processor;
• and
• a memory storing a polarization image processing module, a target scene image enhancement module, and a target scene interpretation module that are programs;
• wherein when the polarization image processing module, the target scene image enhancement module, and the target scene interpretation module are executed by the processor, the polarization intelligent sensing system is caused to:
• perform calculation on the polarization image by the polarization image processing module to obtain polarization information of the target scene;
• generate image information to be restored of the target scene by the target scene image enhancement according to the polarization information of the target scene;
• and
• obtain and output interpretation information including a multi-target detection result of the target scene, by the target scene interpretation module, based on the image information to be restored of the target scene, spectral information, or intensity information through neural networks;
• wherein the polarization optical lens has a weak polarization modulation characteristic;
• wherein the target scene image enhancement module adopts a multi-dimensional target detection neural network based on Detection Transformer (DT), the multi-dimensional target detection neural network comprises a pre-processing sub-module, a transformer sub-module, and a prediction output sub-module;
• the pre-processing sub-module is a feedforward neural network (FNN) configured to take a combination of an intensity image, a polarization degree, a polarization phase angle, a restored target scene image, and the spectral information as an input or a partial combination of the intensity image, the polarization degree, the polarization phase angle, the restored target scene image, the spectral information as the input and perform noise reduction and feature fusion on the polarization image in combination with a noise reduction module and a convolutional neural network (CNN);
• the transformer sub-module is a Transformer model configured to extract related data of a target object;
• and the prediction output sub-module is configured to process data output by the transformer sub-module to obtain the multi-target detection result;
• wherein the multi-target detection result comprises object types and positions of objects in the target scene;
• wherein the target scene image enhancement module adopts a polarization image enhancement model IMG(O)=k1*C1 {I0, I45, I90, I135}+k2*C2 {IMG(D)}+k3*C3 {IMG(A)};
• wherein the IMG(O) denotes the image information to be restored of the target scene, the C1 {I0, I45, I90, I135} denotes a preferred image calculated according to a first polarization image, a second polarization image, a third polarization image, and a fourth polarization image respectively having a linear polarization angle of 0°, 45°, 90°, 135°, the preferred image is a single polarization angle image or a multi-angle calculated image;
• the IMG(D) denotes a polarization degree image, and the C2 {IMG(D)} denotes a first calculation image containing edge highlight information of a target space obtained through the polarization degree image;
• the IMG(A) denotes a polarization phase angle image, and the C3 {IMG(A)} denotes a second calculation image containing surface information of a target object obtained through the polarization phase angle image;
• the k1, the k2, and the k3 respectively denote an intensity coefficient of a calculation preferred image of the polarization image, an intensity coefficient of a polarization degree calculation image of the polarization image, and an intensity coefficient of a polarization phase angle calculation image of the polarization image.