our proposed method for empirical, standardized evaluation of ProApto multispectral camouflage effectiveness

As far as we experienced, the assessment and evaluation of thermal (multispectral) camouflage effectiveness are done mostly by eye. This approach is surely pivotal for assessing how well camouflage works in the Thermal Infra-Red (TIR) spectrum, but we believe that some sort of "visual empirical data" can support the assessment process, also providing guidance for the user.

In this article we describe an easy empirical methodology we developed for objectively and scientifically quantify the degree of thermal camouflage effectiveness and TIR signature management provided by our multispectral camouflage. Our final aim is then to provide our customers with objective data for TIR camouflage effectiveness assessment and ground for camouflage reliability.

The premise revolves around the aim of TIR camouflage, which is to hide the human TIR signature while tuning it in accordance with the surrounding TIR signature. This requires to violate the laws of thermodynamics towards a perverted equilibrium of suppressions and reflections. With this assumption, the ideal expectation for effective TIR camouflage would be to exactly match the average environmental signature, in order to blend in within the surroundings and avoid standing out.

Therefore, we have calculated the average LWIR signature of n.11 LWIR images displaying our TIR camouflage and then we have calculated the difference between the environment and the TIR camouflage.

Long story short, we computed on the basis of Bayesian statistics that ProApto TIR camouflage significantly reduces at short range (30mt - 250mt) human TIR signature regardless of environmental conditions. In comparison with wearing No TIR camouflage, ProApto TIR camouflage can shield and tune human TIR signature around 6 percentual points of difference from the surrounding environment, while not wearing TIR camo results in 44 percentual points of difference on average from the surrounding environment signature. The ProApto TIR camouflage effectiveness is significantly high, as outlined by the statistical analyses we carried out (Bonferroni corrected results, large effect size and high Bayes Factors).

Here below are reported visual summaries of the analyses we carried out (slide the n.11 images below using the arrows). People with knowledge of statistics can scroll down and read in more details methodologies, results and discussions/considerations in "academic fashion".

In the following part of this article we briefly reported the methodological details and the computational apparatus we employed for the image analyses. Despite the fact that this is not the place for open science, we are open to share raw data, codes and proceedings upon reasonable requests.

--------------------------------------------------------------------------------------------------------------------

METHODOLOGY

In order to evaluate how well our TIR camouflage can tune the human signature closely to the environmental TIR signature, we selected 11 long wave infra-red (LWIR) images showcasing our TIR camouflage at close range (from 30mt to 250mt of distance) and we calculated the average TIR signature of the environment as well as the average TIR signature of the operator wearing TIR camouflage.

LWIR images were taken with thermal monocular mounting a 12µm 640x512 LWIR sensor ( NETD ≤35 mk ) during experimental outdoor sessions of TIR camouflage run between Jan 2023 and Feb 2024. The experimental outdoor session were performed in various environments from city to icy alpine and rural areas.

Image processing and calculations

In order to calculate the average LWIR signature of the environment and the LWIR signature of the operators wearing or not wearing TIR camouflage, the n.11 LWIR images were edited as illustrated in Box 1 below. The operators wearing and not wearing TIR camouflage were isolated and removed from the image, and two different images were saved. Then, the operators were replaced with the average image background around their silhouette (this was performed via K-means - nearest neighbor for the surrounding pixels). In this way, the human presence in the LWIR images was removed and the analyses on LWIR environmental signature without human signature "contamination" were made possible.

Box 1 - schematically representing the processing of image analyses employed for the current set of analyses

In Python (version 10.0.19045.4046) a bespoken script was implemented to calculate the brightness percentage and average RGB color code of the edited LWIR images. The script employs the following steps: 1) First, it calculates the mean intensity values of its RGB channels for all the pixels of each edited LWIR image. Then it computes the brightness percentage using the formula (R 299 + G 587 + B * 114) / 1000. This value is then normalized to a scale of 0 to 100 and returned as the brightness percentage of the image. This function also calculates the mean values of its RGB channels, and returns these values as the average RGB color code of the image.

2) Next, the script then calculates the brightness percentages and average RGB color codes for 3 images: the LWIR edited images for environmental signature, the image of the isolated operator with TIR camouflage and the image of the isolated operator without TIR camouflage. So, the script calculates the brightness percentages and average RGB color codes for each of the 3 images using the previously defined functions.

3) Subsequently, it computes the difference between the brightness percentage of the thermal camouflage image and that of the LWIR environmental signature, representing the reduction in detected radiation due to the camouflage.

4) Finally, the script displays the images along with their corresponding brightness percentages and RGB color codes using matplotlib. Each image is shown in a separate subplot with a black background, and the brightness percentage and RGB color code are displayed as titles.

After these computational processes, all the data outcoming from this script and from the experimental sessions were entered in an excel sheet and then analyses in JASP (https://jasp-stats.org/).

Statistical analyses and Bayesian modelling

To quantify the level of TIR camouflage effectiveness and the magnitude of LWIR signature reduction, a Bayesian independent samples T-Test (Mann-Whitney U) was carried out in JASP. The ''operator's LWIR signature'' percentages were treated continuously as a dependent variable while the ''condition'' of wearing or not TIR camouflage was entered as grouping factor (No TIR camo VS ProApto TIR camo).

In a second instance, an ANCOVA model was built in order to calculate whether the LWIR signature reduction of ProApto TIR camouflage was significant in comparison with "No TIR camo". The percentage of discrepancy between the environment LWIR signature and the operators' LWIR signature was defined as dependent variable and the ''condition'' of wearing or not TIR camouflage was treated as a fixed factor. This model was then controlled for the following covariates: distance from the LWIR sensor, air temperature, ground temperature, pressure, relative humidity and wind speed. The model was then corrected for multiple comparisons (Bonferroni correction) and Bayes Factors calculated accordingly.

RESULTS

As visually reported in Box 2, the mean LWIR signature of operators without TIR camouflage is higher than operators wearing ProApto TIR camouflage (mean 69.2 VS 29.3). Performing independent samples T-Test (Mann-Whitney U) resulted in the operator's LWIR signature being significantly managed (p<0.001) by ProApto TIR camouflage, as indicated by the statistical measures of Bayes Factors of 57.52 and Vovk-Sellke maximum p-ratio of 19271. These measures indicate a strong and solid effect - refer to Box 3 for tables.

Box 2 - the upper portion showcases the bar chart of mean LWIR signature difference across the condition of wearing and not wearing ProApto TIR camouflage. The lower portion of Box 2 reports the descriptive statistics for the 11 experimental outdoor session for the Operator's LWIR signature and the Operator VS Environment LWIR signature discrepancy for the condition (No TIR Camo VS ProApto TIR Camo).

Box 3 - the upper portion showcases the Bayesian Independent samples T-Test for mean LWIR signature difference across the condition of wearing and not wearing ProApto TIR camouflage. On the right is plotted the prior and posterior distribution of the Bayesian Modelling. The lower portion of Box 3 reports the T-Test table along with Vovk-Sellke maximum p-ratio.

ANCOVA analysis on the effect of ProApto TIR camouflage in mitigating and tuning the operator's LWIR signature revealed that the Operator VS Environment LWIR signature discrepancy is significantly (p<0.001 Bonferroni corrected) and strongly (BF10 7362; Cohen's D 7.74) managed by ProApto TIR camouflage (see Box 4 for visual rendering and Box 5 for tables). In addition, the effect of ProApto TIR camouflage appears to be independent from environmental conditions as empirically reported in the Bayesian analyses of effect in Box 5, where only the "condition" exhibits a significant Bayes Factor of model inclusion.

Box 4 - the upper portion displays the mean Operator VS Environment LWIR signature discrepancy across the condition of wearing and not wearing ProApto TIR camouflage. The lower portion of Box 4 reports the rainclouds plots along with key statistical measures outcoming from the ANCOVA Bayesian model.

Box 5 - displays ANCOVA tables both for inferential and Bayesian statistics.

Box 6 - variable distribution plots showcasing the distribution of the variables used for the analyses.

CONSIDERATIONS

From these preliminar analyses on ProApto TIR Camouflage effectiveness, we can state that ProApto TIR camouflage significantly mitigates human TIR signature at short range, drastically reducing the signature discrepancy between the environment and the operator (on average around 6% percentual points of difference). The effect of ProApto TIR camouflage is statistically strong and supported both by inferential and Bayesian statistics. In addition, the effect of ProApto TIR camouflage, from the current available data, emerged to be independent from environmental conditions, and this can be therefore generalised as concept of camouflage reliability. However, it should be noted that the current analyses were performed only on n.11 experimental outdoor sessions and more data are needed to properly assess all the aspects of its multispectral effectiveness. Furthermore, several other variables are needed to better control for environmental factors. For example, we did not considered cloud coverage and we did not gather data on global and relative electromagnetic irradiance. This kind of data should be gathered for future and more comprehensive analyses.

FUTURE DIRECTIONS

We are gathering more data and these analyses will be implemented with further experimental sessions. In the following weeks we will release a similar report analysing the conditions where ProApto TIR camouflage did not work. Similarly, another set of analyses will be carried out on the available images of other companies producing Multispectral TIR camouflage, so that we can provide a public ground for camouflage comparison.