Fig. 1. Realization of long-range disorder chiral shells. (a) Schematic diagram of two-step Ag deposition on the microsphere monolayer. The deposition parameters are described by the out-plane deposition angle θ and the in-plane azimuthal angles φ₁ and Δφ. (b) The top-view images of the chiral shell models fabricated with φ₁ = 0°, 15°, 30°, and 45° and Δφ = ±45°. (c) SEM image of the chiral shells fabricated with θ = 60° and Δφ = 45°. The inset image shows the amplified SEM image of the CSs, and the red arrows denote the lattice directions in different micro-domains. (d) The measured transmittance spectra of the RCP (black dotted line) and LCP (red dotted line) lights and the extracted CD spectra (solid pink line). (e) The transmittance of the LH-CSs as a function of the rotation angle of the linearly polarized lights and the wavelength. (f) The LPER spectrum extracted from (e).

Fig. 2. Optical setup and data collection. (a) Schematic diagram of the measured optical setup. The parallel white light illuminates on the blazed gratings, and the diffraction lights on the blazing directions are filtered by one slit to generate monochromatic light. The SoPs of the lights are changed by placing the HWP and the QWP before the CS-covered CCD camera. (b) The micrographs of the CS samples under the illumination of linearly polarized lights with the angles of 0°, 45°, 90°, and 135°. (c) The speckle images recorded by the CCD camera upon the injection of one polarized light. The exposure time is 200 ms. (d) The photoelectric signal intensities from three different pixels as a function of the HWP and the QWP angles at 551 nm.

Fig. 3. ResNet18 model and polarization detection results. (a) Schematic diagram of the CNN ResNet18 model for resolving the SoPs. 1 and 4 represent the numbers of the convolution layer 3 × 3 and 7 × 7 denote the size of the convolutional kernel, and 64, 128, 256, and 512 represent the numbers of the output channels. FC denotes the fully-connected layer. (b) The training and validation loss curves. (c) The predicted (red points) and true (black circles) Stokes parameters plotted on the Poincaré sphere. (d) The polarization detection of the MSEs as a function of the wavelength ranging from 500 to 650 nm.

Fig. 4. Influence factors on the polarization detection precision. (a) The randomness degrees of CD, LPER^-1, and MTD. (b) The SoP number; (c) the pixel number; and (d) the exposure time. The MSE used in (a) is the average MSE of S₁, S₂, and S₃. The wavelength used here is 551 nm.

Fig. 5. Reliability and repeatability of the developed polarimeters. (a)–(b) The true and predicted Stokes parameters for the SoPs (a) included in the reference datasets or (b) not cluded in the reference datasets. The waveplates and grating are not reset in this case. (c)–(d) The true and predicted Stokes parameters for the SoPs (c) included in the reference datasets or (d) not cluded in the reference datasets. The waveplates and gratings are reset in this case. The wavelength used here is 601 nm.