Keywords

1 Introduction

Polymers are an extremely versatile range of materials and have become almost ubiquitous in life of much of the world’s population. Its low cost, high strength, longevity and manipulability make it a primary choice for many applications. However, plastic waste has become one of the most recognized and maligned environmental impacts associated with modern consumerism and has led to a range of policies designed to reduce plastics manufacture and use (e.g. [1, 2]).

Plastic packaging represents a substantial challenge to be addressed; it us responsible for over a quarter of all plastics produced globally [3], and has a short use cycle, after which 95% of its value is lost. Although recycling is an option to retain some of the material value, the European Union has estimated that at end-of-life, less than 30% of plastic waste is recycled, 31% goes to landfill and 39% is incinerated [4] whilst other studies have reported average recycling rates as low as 20% [5]. Resources dedicated to plastics production, transport miles and residual utility are also effectively lost from premature disposal. Within the UK alone, some 2.3 Mt of plastics are used for packaging each year, of which almost 0.5 Mt are consumer pots, tubs and trays (PTT) [6] which are often used to package food products.

An option to reduce plastic waste and its associated environmental impacts is to enable and promote the reuse of polymer products. For the food packaging industry, this requires the need for new circular business models and supply chains, including reverse logistics, cleaning, cleanliness and quality assurance, and both supply chain and consumer incentives [7]. Establishing a circular economy where packaging is returned at its end-of-use, cleaned, and then refilled with new product for resale could provide a solution to resource use and waste production. Food packaging reuse systems already exist (e.g. Loop, Modern Milkman) but packaging materials are typically limited to aluminum and glass. For any reuse system to work for packaging, it is essential to have effective cleaning and quality assurance processes in place.

One technical barrier to the implementation for the circularity of plastic food packaging is the need to effectively clean the packs and provide cleaning assurance to prevent the possibility of product crossover. Such cross-contamination between uses presents the hazard of foodborne illness to consumers which would seriously undermine any potential environmental benefits from such a system. Despite the importance of the cleaning stage, suitable standards, and guidance for cleaning of reusable packing is currently non-existent. It is perhaps difficult to provide such guidance for polymer-based packaging since oft-recommended sanitizing rinse temperatures for food contact surfaces are typically above 70 ℃, which is very near the glass transition temperature of polyethylene terephthalate (PET) and hence may lead to material warping, for example. Also, the added complexity of circular use systems means it could be more difficult to track and control outbreaks of contamination and hence recall dangerous products [8].

The research presented in this paper investigated the feasibility of using ultraviolet (UV) fluorescence imaging to optically detect residual food fouling and thus assuring cleanliness in the case example of margarine tubs. The technique has previously been demonstrated for other food contact surfaces [9], but not for polymer packaging. The paper justifies the use of subject materials before describing the physical and procedural experimental design. Importantly the optical illumination, image capture and image processing are detailed and the results processed and analyzed to provide contextual importance. The optical results are compared against those obtained via an adenosine triphosphate (ATP) swab technique to demonstrate suitability of the process. The UV detection technique is discussed in the context of industrial application within a system for a circular economy for plastic food packaging.

2 Methods and Materials

Within this study two stages of investigation were undertaken to determine, under controlled conditions, the feasibility of using UV fluorescence to detect margarine-like spread fouling on polypropylene (PP) surfaces. The objective was on determine the lowest level of residual fouling that could be detected and to compare this with industrial standards.

2.1 Testing Regime and Sample Preparation

The subject of the studies in this research was Flora® spread, a margarine-type product, retailed in the UK, across Europe and worldwide. Margarine contains ample fluorophores that under suitable excitation of UV light, fluoresce in the green part of the electromagnetic spectrum (~500 to 570 nm) and hence can be imaged. Primarily supplied in PP tubs, the spread was used as an example product that has a high-fat content making it difficult to clean from surfaces and could potentially be supplied in refillable packs. Notably, PP has an advantage for reusable packaging over the more common polyethylene terephthalate (PET) in that it is more resilient to high temperatures required for cleaning.

Critical Detection

The first test conducted sought to determine the critical (minimum) quantity of food residue detectable on a PP substrate. Different dilutions of Flora® spread were created by mixing via agitation with water within beakers placed over a warm water bath. Dilutions were prepared according to those described in Table 1. Using a micropipette, 1 ml of dilution was evenly distributed over a 30 mm diameter test area on uncoloured PP slides of dimension 50 × 50 mm. Samples were left to completely dry in the laboratory ambient atmosphere (typically overnight) before being imaged.

On-Pack Detection

Assessment of fouling levels on an actual pack form was undertaken to determine the suitability of the system for a more realistic assessment scenario. Empty Flora® 500 g PP tubs were utilised in this study. Before each sample preparation, tubs were cleaned manually using detergent and warm water before being rinsed and dried. The inner base and walls of the tubs were individually fouled with 5.0 g, 1.0 g and 0.5 g of Flora® taken from a melted sample and manually distributed near-evenly using a spatula: to provide more realistic distribution of fouling, homogenous distribution was not intended in this study.

Table 1. Solution concentrations investigated. Spread density assumed to be 0.96 g/ml.
Full size table

2.2 Image Acquisition and Processing, and ATP Assay

An optically isolated stainless-steel box was utilized to allow a controlled experimental campaign of digital image acquisition under excitation by UV light, provided via a dual 18W 370nm (nominal) fluorescent lamp (UV18W BLDTU, UV Light Technology Limited, UK). Image acquisition utilised a Basler® (Germany) ace acA1920-150uc 2-megapixel (MP) camera. The camera was mounted in a bespoke holder outside the stainless-steel box at an aperture to ensure stability and positioning. Focus and lens aperture were manually adjusted to optimize image clarity. Once selected, these settings remained constant for the duration of investigation.

Captured images were processed using MATLAB®. The acquired red-green-blue (RGB) image appears as a 2 MP image x 3 elements matrix, where the third dimension represents the three colour channels red, green and blue respectively. Fluorescing fouling appears as a cyan coloured object within the image and can be isolated by extracting the green channel yielding a 2 MP image represented in greyscale.

To distinguish fouling from non-fouled areas, thresholding is required. For this stage, Otsu’s method [10] was investigated and implemented using the ‘multithresh’ function in MATLAB® (MathWorks®, 2021). Otsu’s thresholding method assumes the pixel intensities follow a bi-modal distribution and defines the threshold by minimizing the variance of the values within each of the two classes of pixels it defines. In doing so, the variance between the two classes is maximized [11] are referred to as ‘positive’ and ‘negative’ classifications respectively. Generally and ideally, the number of pixels in the positive area of the image define the amount of detected fouling.

For comparison against an industrially relevant assessment process diluted fouling samples were also assessed using an ATP swab test method (Hygiena® SystemSURE® PlusTM luminometer with UltraSnap® swab). For laboratory practice reasons, separate samples to those investigated via UV fluorescence were prepared for ATP assay albeit with slightly varying dilutions of fouling, but this does not hinder comparison between the techniques. ATP assay is considered the best available technique routinely used in the food industry for evaluating food contact surfaces. Manufacturers guidance for the ATP swabs used recommend a 100 × 100 mm swab area but this was not possible for the concentration studies due to the sample size. In these cases the full slide was swabbed. As recommended by the Hygiena®, swabs were removed from the refrigerator 10 min before being used. For the particular ATP device used the measurement range is 0–9999 RLU (relative light units). A reading of 30 or above is classed as a fail (i.e. unacceptably fouled).

3 Results

For the majority of samples, the image process provided a clear indication of the location of fouling, however noise in the visible part of the spectrum from the reflected illumination source on the base of the dark box made it difficult for the thresholding procedure to differentiate the true fouling from this background signal (see Fig. 1). To address this, images were manually cropped at the border of the sample slide to disregard any area that was not under direct consideration.

Fig. 1.
figure 1

Thresholded image of sample C. The base of the dark box provides false positive pixels.

Full size image

The number of pixels identified through image processing as fouling is shown in Table 2 in comparison to the deposited fouling concentration. The threshold required to differentiate the fouled area from the background was significantly lower for the dilute concentrations (i.e. 9.34 × 10–5 and 4.82 × 10–5) and hence the increased pixel count for the positive regions in these images indicate the difficulty in distinguishing the fouled area from the background noise on the substrate. Hence it is found that the minimum detectable quantity of fouling for this setup and using this image processing procedure lies between 1.74 × 10–4 and 9.34 × 10–5 g/mm2.

The average intensity of the pixels in the positive region of the images also indicates a good sensitivity to concentrations as low as 1.74 × 10–4 g/mm2 (Fig. 2) whilst the low intensities of more dilute fouling suggests the inability of the detection system to reliably differentiate those areas from background noise.

Figure 2 also shows the result of ATP assay for the fouling samples prepared over a similar range of dilutions. In this investigation assessment of fouling via ATP swabbing yields a more gradual relationship with respect to the fouling level and hence could be perceived to be more reliable. However, although not proven conclusively here, the optical detection approach appears to be able to detect fouled/non-fouled regions to the degree required to determine if a surface is ‘clean enough’: a reading of 29 RLU from the ATP assay. Of course, a significant advantage of the optical technique is that it is a non-contact assessment, has no real operational cost and could be performed at high speed in a continuous production environment, unlike ATP assay.

Table 2. Numerical output of image processing for investigated fouling dilutions
Full size table
Fig. 2.
figure 2

The average pixel intensity (blue) from optical processing and ATP response (orange) for a range of Flora® spread dilutions on PP. The orange dashed line represents the pass threshold for ATP assay. The pixel value axis has been scaled to pass the ATP threshold at a similar location to the ATP response for more realistic comparison.

Full size image

The thresholded images of the PP pack fouled with varying amounts of Flora® spread (Fig. 3) have falsely categorized the inside of the dark box (corners of images) as positive for the 1.0 g and 0.5 g fouling levels. This poses the possibility that other areas within the tub may also be categorized as false positives. The tub fouled with 5.0 g of spread shows a more realistic representation of the fouled area in the image but contains negatively classified areas at the corners of the tub base. These are locations where no external printing is present on the semi-transparent polymer. Hence the colour and opacity of the packaging influences the detectability of the fluorescence from the fouling.

Another pertinent observation from the processed images in Fig. 3 is the impact of shadowing on fluorescence of fouling. In the described experimental setup, the camera is positioned directly above the pack forms, but the illumination is offset causing darker regions on one of the longer edges of each tub. Variation in illumination intensity makes a significant difference to the signal obtained from different areas of a fouled surface. The issue is likely to be complicated further by the increased complexity of the pack form. For example, the lids of Flora® tubs, which fasten via an interference fit, have an overhang which makes line-of-sight difficult for optical imaging.

Fig. 3.
figure 3

Raw and processed images obtained from investigating on-pack fouling.

Full size image

4 Concluding Discussion

A new application of UV fluorescence imaging has been demonstrated to be a potentially suitable technique for the application of cleaning assurance for reusable PP food packaging. In its current form the technique performs to a degree of accuracy commensurate with that of the industry standard for food contact surfaces, ATP assay, albeit the potential for some conditions to lead to false classifications needs to be better understood. The optical detection technique has distinct advantages over ATP assay in that it is rapid, non-contact (hence does not contaminate the surface), has negligible operational costs and does not produce consumable waste (swabs). There are however several potential improvements that should be made to improve sensitivity, reliability and suitability for direct industrial application.

On the one hand, improvements to the detection system such as increased and more homogeneous light intensity could yield improved images, as could a detection system with improved quantum efficiency (for the current camera this is 54%) and filtering to improve signal-to-noise. In terms of image processing Otsu’s thresholding method is highly sensitive to distributions of intensities and hence it is possible to fail to detect small areas or low levels of fouling. A two-stage process may be better: rapid detection for obvious fouling, then a different process, for example hysteresis thresholding, to detect lower levels of fouling. Intelligent definition of the region of interest to determine the location and orientation of packs would remove the likelihood of false positive classification from irrelevant areas of the image.

With respect to packaging design to support detection of cleaning success (and also improve the cleaning process), complex geometries, particularly partially enclosed spaces and tight corners, should be avoided. Wide and shallow containers with low draft angles are preferable for improved visibility and to support more uniform illumination. These practices should be shared across all packaging suppliers and indeed the Ellen MacArthur Foundation [12] has suggested that reusable packaging should be standardized to allow for better system design and reliability.

A circular system for polymer packaging is a potential solution to plastic consumption but demands changes to current industrial supply chains. Packaging manufacturers would need to become packaging system providers, onboarding reverse logistics, cleaning, and quality assurance processes. The cleanliness assurance method investigated in this work, although not fully developed, indicates that it could form part of the technological solution to reusable plastic packaging. Increasing assurance of food safety is likely to remove one barrier to consumer and industrial acceptance of such a radical and beneficial transition.