A Recongurable Perovskite/Polymer Composite Film for Real-Time Detection of Agricultural Spraying

Responsive composites that can display sophisticated responses under environmental stimuli are of paramount importance for developing smart materials and systems. However, the hierarchical design of their multiscale constituents to achieve such response remains a challenge. Here, we report a responsive polymer composite obtained by integrating hierarchical interactions between the polymer network meshes, perovskite nanoinclusion, and a microstructured layout. More specic, a layered composite lm has been made with perovskite nanoparticles embedded in a hydratable polymer network as the top layer. The perovskites inclusions can undergo a reversible transformation between a nanocrystalline state and a dissociated ion state, triggered by spraying aqueous solutions on the polymer top layer, resulting in an on/off switch of uorescence at 510 nm. Meanwhile, the surface layer experiences a recongurable micro-wrinkling that can gradually change the lm transmittance between 90% and 10%. The two orthogonal responses show a good reversibility for at least 15 cycles. They can be manipulated independently as they respond differently to the amount of water applied. We demonstrate the use of such lm by real-time, quantitative, and repeatable detection of spraying and subsequent droplet distribution. Such a sensing capability is urgently needed in precision agriculture for fast assessing the deposition quality of pesticides and fertilizers, yet still not available. Our ndings enable the design of perovskite-based responsive composites with multiple functions as well as novel device applications in sensors, actuators, and optoelectronics.

can be utilized and manipulated with a hydrophilic and hydratable polymer network, leading to novel responsive polymer composites with hierarchical synergies and exotic functionalities capable of complex device operations.
We rst designed the structure and fabrication scheme of a multi-layered and perovskite-embedded composite lm (Fig. 1a, also see details in "Methods" and Figure S1). A polymeric three-layer lm was constructed via stepwise spin-coating (Fig. 1a, left). The top layer is a water-absorbing polyvinyl alcohol (PVA) lm that was cross-linked with sodium benzoate under light irradiation ( Figure S2). This layer can be arbitrarily patterned by using masks (see Figure S3 for arraying process); the middle layer is a compliant polydimethylsiloxane (PDMS) lm; and the under layer is a transparent supporting substrate based on polyethylene terephthalate (PET), polymethyl methacrylate (PMMA) or glass. A ligand-free inorganic cesium lead halide (Cs-Pb-Br) perovskite was next introduced into the cross-linked PVA layer by swelling it with the precursor solution (Fig. 1a, middle), and the perovskite content was regulated by tuning the solution concentration ( Figure S4). After subsequent drying and post-treatment, we obtained the desired composite lm (Fig. 1a, right). Cross-sectional scanning electron microscope (SEM) images ( Fig. 1b and Figure S5) and corresponding energy dispersive spectroscopy (EDS) results ( Figure S6) of the as-prepared lm show that perovskite nanoparticles (NPs) with sizes of 200-400 nm are evenly distributed in PVA (also see uorescence micrograph in Figure S7 and transmission electron microscope images in Figure S8).
The prepared composite lm is transparent under natural light and uorescent under ultraviolet (UV) light ( Fig. 1c, left). After wetting with an ultrasonic humidi er ( Figure S9), the lm becomes translucent and non-uorescent (Fig. 1c, right). Meanwhile, the patterned lm with arrays of circular disks (Fig. 1d, left and middle) showed a similar change of the uorescence and transparency upon water spraying (Fig. 1d, right and Figure S10). In particular, photoluminescence (PL) spectroscopy showed that the emission peak of the lm in the dry state locates at around 510 nm, with a full width at half maximum (FWHM) of about 20 nm. After wetting, the emission peak completely disappeared (Fig. 1e). This on/off switching of uorescence is reversible and reproducible (see Video S1) with a consistent peak position and intensity for at least fteen cycles (Fig. 2f). We also found that the transmittance of the lm in the visible light range was above 90 % in the dry state and below 10 % in the wet state (Fig. 1g). This is caused by the wrinkle formation in the swollen PVA layer that can signi cantly scatter incident light 46 (bottom inset in Fig. 1g, and see the 3D microscope image in Figure S11). The changes in transmittances also demonstrated good reversibility and repeatability (see Video S2) in a fteen-cycle study (Fig. 1h), as the PVA layer becomes at again after de-swelling (top inset in Fig. 1g).
To understand the reversible uorescence change, we conducted X-ray photoelectron spectroscopy (XPS) and X-ray diffraction (XRD) analysis of the composite lm. In the dry (original) state. XPS results ( Figure  S12 and Table S1) show that the embedded perovskite has a stoichiometric composition of CsPb 2 Br 5 .
The XRD study presents characteristic peaks at 11.7°, 23.4°, 35.4° and 47.8°, tting well with the crystal planes (002), (210), (312) and (420) of the tetragonal CsPb 2 Br 5 phase 47 (PDF#25-0211), respectively ( Fig. 2a and Figure S13). These peaks disappeared after water treatment and recovered after drying ( Fig. 2a). We thus assume that perovskite NPs have dissociated into free ions (Cs + , Pb 2+ , Br − or PbBr 6 4− ) that are dissolved in the swollen PVA network in the wet state, and the ions can recrystallize into NPs again after water evaporation (Fig. 2b). To verify this, electrical measurements on the same lm sample were performed ( Fig. 2c and Figure S14). The conductivity of the lm indeed increases from 10 − 4 S·m − 1 to 10 − 1 S·m − 1 after wetting and reduces back to 10 − 4 S·m − 1 after drying. The conductivity in the wet state is at least one order of magnitude higher than that of the reference sample containing no perovskite. These results further explain the uorescence quenching in the wet state and subsequent recovery in the dry state.
The reversible dissociation and recrystallization of perovskite species above is highly intriguing in connection to fact that perovskites are highly susceptible to water molecules, i.e. water would usually induce irreversible dissociation of these ionic crystals, strongly affecting their opto-electric propertiese 48-50 . Such a reversible ion-to-crystal transformation (RICT) can be attributed to the cross-linked PVA network. Further control experiments show that perovskite NPs cannot grow uniformly in PVA lm without cross-linking ( Figure S15). Likewise, their dissociation becomes irreversible without cross-linking. Notably, PVA hydrogels have been used as solar steam generators in seawater desalination for restricting ion diffusion while allowing water evaporation 51,52 . Their networks can also be used as templates to regulate the growth of metal NPs such as Fe, Co, Ni, Cu, etc. 53,54 Thus, we postulate that the PVA network serve as semi-enclosed "cages" (Fig. 2b) that not only absorb a su cient amount of water to dissociate perovskite NPs but at the same time con nes the migration of ions (Cs + , Pb 2+ , Br − or PbBr 6 4− ) and subsequently nano-spatially nucleates their recrystallization for realizing RICT. It is worth noting that such RICT also works for a lead-free double perovskite Cs 2 AgBiBr 6 ( Figure S16), i.e., its composite lm also shows good reversibility and reproducibility of the uorescence signal in at least four cycles ( Figure S17).
We further studied the sequence of wrinkle formation and transformation of perovskite NPs and their eventual synergy during wetting. Our real-time monitoring of water spraying on the composite lm shows that the transparency declined (1-2 s, Video S2) before the uorescence quenching takes place (4-5 s, Video S1). This result suggests that water molecules rst permeate and swell the PVA matrix to form wrinkles and subsequently react with the embedded perovskites NPs to dissolve them into the constituting ions. We further prepared different composite lm with different amounts of perovskite (2.5-11.5 vol.%) with a xed PVA layer thickness of 3.6 µm and measured the water absorption amount per unit area of these lms. We found that the amount of water required to completely quench the uorescence increases from 3.2 to 4.5 mg/cm 2 as function of increasing perovskite concentration, while the amount of water needed to form wrinkles remains almost constant at a value of ~ 0.6 mg/cm 2 (Fig. 2d). These results con rm that water rst triggers wrinkles and then quenches the uorescence.
Interestingly, we could quantitatively evaluate the amount of water absorption based on the light scattering properties of the swollen lms. In particular, we built a measuring setup by combining a camera and a precision balance to synchronously record the surface brightness (grayscale) and weight of a lm sample in real time. (see details in "Methods" and Figure S18). We could thus obtain time-varying curves of the average grayscale value and water absorption (Fig. 2e) and quantitatively establish their relation after normalization with a reference sample ( Fig. 2f and Figure S19). For the lm sample with PVA thickness of 1.8 µm (6 wt.% cross-linker), a linear dependency in the range of 0-0.3 mg/cm 2 can be observed, with good reproducibility in repeating runs, even at different illuminances (15 and 120 lux). Moreover, this linear response range can be precisely engineered by tuning the lm thickness ( Figure  S20 and S21). Speci cally, increasing the PVA layer thickness from 1.8 to 11.6 µm linearly increases the saturated absorption of water from 0.3 to 2.4 mg/cm 2 (Fig. 2g), respectively, approaching the uorescent sensing range of bespoke perovskite NPs.
Based on the above results, we anticipate the prospect to regulate the water absorption capability (or sensing threshold) and linear response range of the composite lms by tuning the perovskite content and PVA thickness, respectively. Notably, the on/off switching of uorescence is expected not to interfere with the gradual change of transmittance or grayscale of the lm, and these two responses can thus be manipulated independently as the uorescence switching of perovskite requires a large amount of water qualitatively, while wrinkle formation is a gradual physical process sensitive to a lower amount of water.
This markedly different response of the different constituents of the polymer composite will be critical for practical applications in sensing and other elds.
As a demonstration, we studied the application of the composite lm as a sensor in the real-time detection of water spraying. Rapid, low-cost, and non-disposable detection of the amount and distribution of sprayed droplets at given locations, e.g., on the leaves or in the eld, is urgently needed in precision agriculture but yet to be realized. Currently, water sensitive paper (WSP) is widely used for the agricultural spray detection [55][56][57][58] , which is neither reusable nor suited for real-time monitoring. Moreover, it follows a routine of " eld sampling rst and lab analysis later" that severely limits its application in today's unmanned agricultural practices.
We rst conducted a droplet impact experiment and found that our composite lm with hydrophilic surface facilitates the droplet deposition with little rebounding or splashing ( Figure S22). Ideally, we can obtain real-time spraying data by recording the distribution of grayscale values or uorescence at different positions of the composite lm. In practice, however, we found that water droplets could spread laterally on the surface layer and overlap with each other, resulting in a reduced spatial resolution (Fig. 3a,  left). To address this issue, we used the patterned composite lm containing arrays of circular disks (~ 300 µm) with the assumption that the isolated cells can work independently to localize the diffusion of water (Fig. 1d). We found the arrayed lm provided a much higher resolution of the droplet distribution under the same spraying conditions (Fig. 3a, right), i.e., the response of the spots remained small in size and was almost free of lateral extension (also see Figure S23). We further took a series of micrographs from a typical spray-recovery process (Fig. 3b, also see Video S3). As expected, the hydrophobic PDMS patches in between cells prevents water from spreading, and the distribution of the wrinkled cells re ects the distribution of water droplets. We thus can obtain the distribution of the sprayed droplets by using the patterned lm sensor.
Moreover, we found that the patterned structure also signi cantly improves the overall exibility of the lm, enabling it to be arbitrarily folded or twisted with little effect on its sensing capability ( Fig. 3c and Figure S24). A bending test under a microscope showed that at the same bending radius of 2.5 mm ( Figure S25), the continuous lm without array presented unfavorable wrinkles and even cracks (Fig. 3d,  left), while the curved lm with arrays retained the original intact morphology (Fig. 3d, right). The arrays break the continuity of the surface layer, which reduces the shear stress transfer between layers during bending and thus minimizes mechanical strain and wrinkles [59][60][61][62] . These results highlight the application potential of our composite lm in wearable detection.
We next studied the sensing performance of our composite lm in different application scenarios. We rst demonstrated a bandage-like device that can be attached onto plant leaves (Fig. 4a). In particular, we peeled off the pre-prepared composite lm (~ 5 vol.% perovskite in 3.6 µm PVA on PDMS, size ~ 1×1 cm) from its original substrate (glass) and transferred it to leaves of a plant (Sche era octophylla, for example). The exible lm can be readily tted to the leaf surface (Fig. 4b). Its uorescence disappears after spraying and recovers after drying, ready for the next test (Fig. 4c). Therefore, we consider that the composite lm can be used as a wearable sensor in practice, i.e., by placing it at different positions of the crop (leaves, stems, fruits, etc.) for the on/off detection of spraying and its distribution. Notably, the response range of our composite lm to water (0-4.5 mg/cm 2 , Fig. 2d) is consistent with the actual spraying densities of pesticides 63,64 (e.g., insecticides, herbicides, fungicides, etc.) ranging from 0.5 to 3.0 mg/cm 2 (or 50-300 L/ha) in agriculture. Moreover, uorescence-based detection, enabled by our composite lm, can be performed under low ambient light or even at night.
We further designed a portable device (Fig. 4d) with the composite lm sitting on the top of a sealed home-made box for spraying detection in the eld. A micro-camera with Wi-Fi transmitter was installed at the bottom center of the box for monitoring the lm and transmitting real-time signals wirelessly. A water spraying test with this device was rst performed in a laboratory environment, acquiring the result of one spray-recovery circle ( Fig. 4e and Figure S26), where the white spots were attributed to the local wrinkles triggered by water droplets. After the test, our device could recover spontaneously and be reused. We further performed an outdoor test in a eld environment under natural, ambient light ( Figure S27) and obtained results after background subtracting (Fig. 4f). A database of droplets could be established based on image recognition, and we could thus statistically analyze and acquire the desired spraying information quantitatively including the number of droplets (N), their average size (s), the average spacing between droplets (d), the total water volume (V), and the coverage ratio (C) (see the analysis and a typical list of results in Figure S28). Moreover, because the sensor devices are portable and recoverable, they can be placed at different positions (p 1 , p 2 , ...p n ) and record a series of data at different moments (t 1 , t 2 , ...t n ), respectively. All these data obtained from our composite lm sensors are acquired in real time, and thus they can provide timely feedback and guidance on eld operations, e.g. spraying operation by unmanned aerial vehicle (UAV). In this way, we can achieve precise spraying control of pesticides in an automated and intelligent manner for precision agriculture (Fig. 4g).
In summary, we have developed recon gurable polymer composites presenting water-switchable transmittance and uorescence with good reversibility and independent controllability. The top hydratable layer of the composite can form surface wrinkles reversibly, triggered by water, while perovskites inside the layer undergo reversible ion-to-crystal transformations because of nanocon nement in the voids of the crosslinked hydrogel. The two responses can work cooperatively, permitting an unrivaled sensing application of the composite lm in precision agriculture, i.e., a rapid and reusable on-site detection of water spraying in terms of quantity, threshold, and distribution. Our ndings suggest that multiple functionalities can emerge from the hierarchical interactions between the multiscale constituents, which would give access to perovskite-based smart composites and exible devices with novel applications.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. VideoS3.mp4 VideoS1.mp4 VideoS2.mp4 Supportinformation.docx