Surface plasmon resonance microscopy (SPRM) is a near-field sensing and imaging technique that is able to detect tiny changes in object properties within plasmon wave penetration depth with prominently high sensitivity [1]. It provides surface plasmon resonance (SPR) imaging of samples in the near-field region for optical visualization and sensing, which is highly desirable in chemical and biological studies. Due to its high detection sensitivity, high-throughput, nondestructive, noninvasive, wide-field, and label-free imaging advantages, SPRM has excellent capabilities for optical sensing measurements and microscopic imaging and has been widely employed for cellular activities [2–5], antigen–antibody binding processes [6–9], and biomolecular detection [10–13], as well as the characterization of the physical properties of nanoparticles [14–17] and two-dimension materials [18–21]. SPRM is typically constructed by the Kretschmann configuration with a prism or a high numerical aperture (NA) objective. The Kretschmann configuration is the most common SPR excitation configuration, in which a thin metal film is evaporated on top of a glass substrate. Prism-coupled SPRM has a large field of view but low imaging resolution because it cannot be integrated with high magnification objectives due to the limitations of the system structure. While utilizing a high NA and magnification objective, the objective-coupled SPRM effectively improves imaging resolution and avoids imaging aberrations [22].

However, the propagating surface plasmon wave (SPW) is generated at the interface of thin metallic film while SPR occurs. When the SPW interacts with subwavelength objects on the metallic interface, scattering light arises from the interactions and interferes with the reflected light in the imaging optical path since they are coherent, giving tail-shaped interference patterns in the SPR images [23]. This causes an unclear edge of the sample along the direction of SPW propagation and seriously deteriorates the image contrast, which significantly hinders the broad applications of SPRM. The demands for full morphology visualization and higher image contrast have driven research into more advanced approaches to improve SPRM imaging capabilities. The current related approaches can be categorized into four types. The first one is limiting the propagation length of SPW to suppress the “tail” patterns, e.g., by optimizing the light wavelength [24], replacing the metallic film with plasmonically active metamaterials to excite localized SPR [25], upgrading the flattened gold substrate to a grating nanostructure to excite the “hybrid” SPR modes [26], and utilizing high-Q dielectric metasurfaces instead of plasmonic configurations to avoid the generation of SPW [27–29]. Due to the complexity of the optimization and nanofabrication process, these methods are difficult for wide applications. The second approach is acquiring multiple images at different azimuthal incidence angles followed by image fusion algorithms to eliminate the “tail” patterns, such as by switching opposite illumination [30,31], dual-channel simultaneous illumination [32,33], multi-angle illumination [34–36], and rotating the sample [37]. However, this type of method requires several tunable mirrors or galvanometers, making the design of the optical system cumbersome. The third approach is utilizing image processing algorithms to enhance the SPRM imaging capabilities with a single image, such as deconvolution algorithm [8,38] and spectral filtering with a designed mask [39]. Currently, such methods are only applicable to simple samples such as nanospheres and nanowires. The fourth approach employs advanced optoelectronic devices, such as a digital light projector [40] and a hyperspectral microscope [41], to improve the SPRM image quality but it comes at a high system cost.

Here, we propose an incidence angle scanning method in an objective-coupled SPRM to improve the imaging capabilities. By scanning a wide range of incidence angles, we collect a series of intensity images to build a data cube. Then we use the designed algorithm to extract the resonance angle corresponding to the minimum value of the reflectance pixel-by-pixel, and further reconstruct the resonance angle (RA) image in which “tail” patterns and the background noise have been prominently suppressed. As a result, this method is able to reconstruct the complete morphology of samples and provide exceptional imaging contrast. Our method can be easily integrated into the existing SPRM to enhance the imaging capabilities, while possessing the advantages of the high detection sensitivity of SPR. The validity of this method has been demonstrated by imaging photoresist square samples and ReS2 two-dimensional materials in comparison to conventional SPRM.

The regions above the metallic film interface with different physical parameters, such as varying dielectric constants and thicknesses, require different incidence angles to excite SPR. In other words, at a specific incidence angle, only the regions with the same physical parameters are at the SPR excitation state, leading to the minimum reflectance. Therefore, the regions with different physical parameters will resonate separately by gradually adjusting the incidence angle of the excitation light. During this process, a series of intensity images are recorded. Then, the RA values of all regions are extracted pixel-by-pixel. By relabeling the image with the acquired RA data, the sample morphology can be fully recovered and effectively differentiated from the background, resulting in the superior enhancement of the SPR image contrast. Figure 1 illustrates the workflow of the proposed method.

The experimental setup of the objective-coupled incidence angle scanning SPRM is shown in Fig. 2. A monochromatic linearly polarized laser beam from a stabilized diode laser (center wavelength 632.8 nm, 0633-08-11-0030-100, Cobolt, Solna, Sweden) passes through a negative lens to be expanded and then is collimated by a collimation lens. A focusing lens is employed to focus the beam at the back focal plane of the oil immersion microscope objective (100×, 1.49 NA, Nikon, Japan). A plane wave output from the objective then illuminates the Kretschmann configuration in a wide-field manner at incidence angle θ. A half-wave plate is inserted between the focusing lens and objective to generate p-polarization light to excite SPR. The reflected beam from the Kretschmann configuration carrying the sample information is collected by a tube lens and imaged on the CCD (Basler ace acA2040-90um, Ahrensburg, Germany). The fiber output, negative lens, collimation lens, focusing lens, and half-wave plate are mounted on a linear translation stage to adjust the offset between the laser beam and the optical axis of objective (inset of Fig. 2). By controlling the translation stage to linearly move forward step by step we can perform θ scanning in a wide range. The angular resolution of the experimental system depends on the stepping resolution of the linear translation stage. The unidirectional repeatability of the used linear translation stage is 10 μm and the corresponding angular resolution is 0.43° (see Appendix A for details). The detection limit of the sample refractive index variation is 0.0072 corresponding to the angular resolution (see Appendix B for details). We further calculate the correlation between θ and the translation stage position d with the formula d−d0=fsinθ, where d0 and f represent the original translation stage position and the equivalent focal length of the objective, respectively. SPR intensity images at different incidence angles are captured as the raw data for further image processing.

The proposed data processing procedure is shown in Fig. 3. First, raw data including SPR intensity images and background images at different incidence angles are captured using the mentioned angle scanning mechanism. Liquid dielectric material such as water is added above the samples to disrupt the resonance conditions to obtain the background images. Then, background subtraction and Fourier space filtering are performed to remove system background noise and interference noise. The processed image data are combined to form a data cube D(x,y,θ) with size of s1×s2×n, where s1, s2 denote the size of each image and n is the number of incidence angles. θ needs to satisfy {minΘ−α≤θ≤maxΘ+βΘ={θSPR,i,1≤i≤m}α,β∈R,(1)where m denotes the number of sample types, Θ represents the aggregation of all SPR excitation angles, and the biases α, β are used to restrict the upper and lower limits of θ.

The mathematical principle of the designed algorithm is expressed by {R=D(xi,yj,θ),1≤i≤s1,1≤j≤s2I(xi,yj)=θSPR,R(θSPR)=minR,(2)where R denotes the reflectance. For each pixel (xi,yj) in the cross-section of the data cube, R is extracted for all incidence angles along the longitudinal dimension of the data cube. Through searching the minimum R, the corresponding angle (e.g., resonance angle θSPR) is obtained and assigned to matrix I(xi,yj). Finally, all pixel locations are scanned to obtain the image I(x,y).

Since the value at each pixel of the image represents the resonance angle, we name the calculated image as the RA image. In fact, the RA image is constructed using the SPR angle information. Since the resonance angles for different positions of the sample are different, the RA image can distinguish different parts of the sample clearly with high contrast. Compared to a conventional SPR image, there are no “tail” patterns in the RA image, the sample edges are sharper, and the morphology of the sample is fully recovered. Due to the pixel-by-pixel calculation manner, each pixel generates an SPR curve. Therefore, effects of coherent illumination, such as intensity variations and interferences, can be significantly reduced, resulting in a remarkably increased imaging contrast.

However, when SPW propagates from the resonant region to the nonresonant one, interference is generated [42] and the dark areas of interference patterns affect the search of minimum R, resulting in inaccurate resonance angle calculation and the loss of sample edge information. Therefore, once the RA image is obtained, we further assess the imaging quality based on the sample morphology fidelity and background noise level. Sample morphology fidelity refers to the matching degree between the sample morphology of the RA image and that of the optical microscope image. If the sample edge information is partly lost or the background noise is still obvious, the calibration algorithm is utilized for further correction. We propose to regenerate two RA images [i.e., RA images with small (SRA) and large (LRA) scanning ranges]. For the case of SRA, the incidence angle scanning range does not cover the resonance angle of the background region. In this way, no interference fringes are caused by the cross-region propagation of SPW and thus the SRA image contains complete sample morphology. On the contrary, if the resonance angle of the background region is covered, the generated LRA image has a more uniform and smoother background. A level-set image segmentation algorithm is a digital method for boundary extraction by tracking contours and surface motion that is widely applied for image post-processing. Thus, we apply the level-set segmentation algorithm on the SRA image to produce a mask and an anti-mask representing the sample and background regions, respectively. We use the mask to extract the sample information from the SRA image and the anti-mask to extract the background information from the LRA image. Finally, the extracted information is combined to build a calibrated RA (CRA) image with full sample morphology information and increased image contrast. Note that the large scanning range actually includes the small scanning range, so only the large-range scanning is performed during the experimental data collection.

We experimentally verified the feasibility of the incidence angle scanning SPRM by using photoresist square samples with different side lengths as the imaging targets. Photoresist samples on the gold film were fabricated through spinning coating and electron beam lithography. First, we performed incidence angle scanning (the scanning range was set as 40.9°–63.0°) by controlling the linear translation stage to step forward with an interval of 10 μm and recorded a series of SPR intensity images. A total of 71 images were collected and the time consumption was about 14.5 s. It is important to note that the quality of the image reconstruction is directly affected by the number of images collected during the scanning process (see Appendix C for details). Then, the RA image was obtained using the image processing procedure described above. The interference patterns generated by the cross-region propagation of SPW from the photoresist-covered area to the air area seriously hindered the image reconstruction, which caused partial loss of the square edge information in one direction of the initial RA image. Therefore, further corrections for all photoresist square samples with different side lengths were needed to obtain the CRA images by applying the calibration algorithm.

Figures 4(a₁)–4(a₃) show the SPR image in which SPR occurred in the photoresist-covered area, the CRA image, and the bright-field image recorded by a commercial reflected light microscope. It can be seen that the sample edge in the y direction, which was the SPW propagation direction, was completely lost in the SPR image [Fig. 4(a₁)]; at the same time, the edge information was fully recovered in the CRA image of Fig. 4(a₂). The geometric shape of the sample in the CRA image conforms well with the bright-field image [Fig. 4(a₃)], confirming that the CRA image can reflect the sample geometric morphology as reliably as the microscope image. In addition, the contrast of the CRA image is much higher than that of the SPR image. For a further detailed comparison, the cross-section line-cut profiles along the x axis and y axis were extracted from the SPR and CRA images, respectively, as shown in Fig. 4(a₄). All of the profile plots were assigned to square sample areas with an actual size of 15μm×15μm. The profiles of the CRA image exhibit abrupt changes at the edges in both the x and y directions, while the SPR image profiles show gradients and perturbations in the x direction and barely any edges in the y direction. The differences between the measured sizes based on the CRA image and actual sizes in the x and y directions are 290 nm and 90 nm, respectively, which means that the method allows for precise two-dimensional quantification measurement.

The experiment results of the photoresist square samples with the side length of 8 μm are shown in Fig. 4(b). It can be seen that the sample morphology in the y direction of the SPR image can no longer be effectively distinguished as two neighboring squares. It can, however, be easily identified in the CRA image, which is consistent with the bright-field image. It should be noted that due to the limitation of the image segmentation algorithm, sample boundaries cannot be always accurately extracted. So, there is an edge-bending problem in the CRA image and sometimes the reconstructed square is not the standard square. In addition, some speckle noises are still present in the CRA image, which may be caused by the combination effect of the “tail” patterns and inherent noises from the coherent light illumination. A higher accuracy of image segmentation is foreseen to make use of more sophisticated neural network-based image segmentation methods and machine learning.

To quantitatively compare the image contrast, the contrast-to-noise ratio (CNR) is adopted as the figure-of-merit and defined as [43] CNR=1M∑SsigI−1N∑SbackIσsys,(3)where I is the value of the pixels, Ssig is the calculated signal area, Sback is the background area, M and N denote the number of pixels in the corresponding area, and σsys represents the standard deviation of system noise. Based on this formulation, the CNRs of the CRA images and corresponding SPR images for photoresist square samples with different side lengths were calculated and are shown in Fig. 5. These results indicate that a significant improvement in the image contrast has been achieved in all CRA images, with an average fivefold increase.

To further verify the feasibility of the proposed method, we conducted an experimental study on ReS2 two-dimensional (2D) materials. The ReS2 materials were transferred onto the Au substrate with the mechanical stripping and dry transfer process. The data collection process was the same as the photoresist square sample. The results obtained by using incidence angle scanning SPRM are presented in Fig. 6. Since the thickness of the ReS2 2D materials is only a few dozen nanometers, as proven in the atomic force microscope (AFM) measurement results in Fig. 6(d), the disruption induced by the cross-section propagation of SPW is limited. This enables direct reconstruction of the high-contrast RA image with complete morphology of the materials without applying the calibration algorithm. The SPR image shown in Fig. 6(b) is unable to distinguish the edge of the ReS2 material along the propagation direction of SPW, while the RA image in Fig. 6(c) has clearly distinguishable edges and conforms well with the bright-field image [Fig. 6(a)]. The edge sharpness and quantification measurement of the materials can be further investigated from the cross-section line-cuts of the images. From the profile distributions in Fig. 6(e), no edge can be found in the SPR image, but there is a clear jump at the edges in the RA image. The lengths of the long and short sides were measured to be 3.963 μm and 2.426 μm, and the difference between the measurements and AFM results is 53 nm and 120 nm, respectively. We further quantitatively evaluated the image contrast with the CNR metric for the ReS2 2D materials, as depicted in Fig. 6(f). The results show that the RA images have an average twofold higher CNR compared to the SPR images.

The experiment results above demonstrate that the RA image obtained from incidence angle scanning SPRM with the designed data processing procedure is superior to the grayscale SPR image in terms of the image quality, especially for sharpening the edges of the samples. It provides the complete morphology of the samples with exceptional image contrast. Besides, we can perform precise 2D quantification measurements using the RA image, which is incapable in conventional SPRM.

It is worth mentioning that not all samples are applicable to this method. Samples that are compatible with this method should feature the following characteristics. First, SPR can be excited in the sample area within the scanning range of incidence angle. Second, the sample should possess a certain surface area when adhered to the gold film. Besides, all of these factors have an impact on CNR calculation: the surface flatness of the coverslip and the gold film fabricated with the E-beam evaporator, the etching accuracy of the electron beam lithography for photoresist sample fabrication, and the thickness and shape of the two-dimensional material and its adhesion degree to the gold film. Therefore, the CNRs of samples with different shapes or diverse sizes show discrepancies.

In conclusion, we have proposed and demonstrated a flexible incidence angle scanning SPRM for enhanced SPR imaging. With photoresist square samples and ReS2 samples, we have experimentally demonstrated that the RA image obtained by the proposed data processing procedure can fully reconstruct the sample morphology, prominently eliminate the influence of SPW propagation, and significantly improve the image contrast. We have evaluated the improvement of the image contrast quantitatively using a defined metric contrast-to-noise ratio and the results indicate that a maximum fivefold increase can be achieved. Besides, we can perform precise 2D measurements with an error margin of hundred nanometers, which is incapable in conventional SPRM. Compared to conventional SPRM, our method operates in an angular scanning mode and has flexible features. First, it can provide two types of images, including a grayscale SPR image and the RA image. Second, it enables full morphology imaging with high contrast based on single-pixel angular scanning. Third, it can be used to measure the 2D profiles of dielectric samples using the RA image. Fourth, because the setup is almost the same as conventional SPRM, the method can be easily migrated to upgrade existing SPRM with low cost and stable performance. With the remarkable features and excellent performance mentioned above, our method will promote the application of SPRM in label-free detection, imaging, and quantification of chemical and biological measurands.