Data-Centric Performance Improvement Strategies for Few-Shot Classification of Chemical Sensor Data

Metal oxide (MOX) sensors offer a low-cost solution to detect volatile organic compound (VOC) mixtures. However, their operation involves time-consuming heating cycles, leading to a slower data collection and data classification process. This work introduces a few-shot learning approach that promotes rapid classification. In this approach, a model trained on several base classes is fine-tuned to recognize a novel class using a small number (n = 5, 25, 50 and 75) of randomly selected novel class measurements/shots. The used dataset comprises MOX sensor measurements of four different juices (apple, orange, currant and multivitamin) and air, collected over 10-minute phases using a pulse heater signal. While high average accuracy of 82.46 is obtained for five-class classification using 75 shots, the model’s performance depends on the juice type. One-shot validation showed that not all measurements within a phase are representative, necessitating careful shot selection to achieve high classification accuracy. Error analysis revealed contamination of some measurements by the previously measured juice, a characteristic of MOX sensor data that is often overlooked and equivalent to mislabeling. Three strategies are adopted to overcome this: (E1) and (E2) fine-tuning after dropping initial/final measurements and the first half of each phase, respectively, (E3) pretraining with data from the second half of each phase. Results show that each of the strategies performs best for a specific number of shots. E3 results in the highest performance for five-shot learning (accuracy 63.69), whereas E2 yields the best results for 25-/50-shot learning (accuracies 79/87.1) and E1 predicts best for 75-shot learning (accuracy 88.6). Error analysis also showed that, for all strategies, more than 50% of air misclassifications resulted from contamination, but E1 was affected the least. This work demonstrates how strongly data quality can affect prediction performance, especially for few-shot classification methods, and that a data-centric approach can improve the results.

Reproductivity Study of Metal Oxide Gas Sensors Using Two Different Temperature Setups

The use of the electronic nose as a screening device is of great interest in various types of applications, including food quality control and environmental monitoring. It is an easy-to-use device and produces a much faster response than that obtained by classical chemical and microbiological techniques. The reproductivity of nominally identical electronic noses and sensors is critical. Four identical MOX sensors were compared using two different working methods, namely, the temperature modulation mode and isothermal mode. Each sensor was tested with two standard compounds, water and lactic acid, often identified in food matrices, which are potential applications of the electronic nose.

Tin, Titanium, Tantalum, Vanadium and Niobium Oxide Based Sensors to Detect Colorectal Cancer Exhalations in Blood Samples

User-friendly, low-cost equipment for preventive screening of severe or deadly pathologies are one of the most sought devices by the National Health Services, as they allow early disease detection and treatment, often avoiding its degeneration. In recent years more and more research groups are developing devices aimed at these goals employing gas sensors. Here, nanostructured chemoresistive metal oxide (MOX) sensors were employed in a patented prototype aimed to detect volatile organic compounds (VOCs), exhaled by blood samples collected from patients affected by colorectal cancer and from healthy subjects as a control. Four sensors, carefully selected after many years of laboratory tests on biological samples (cultured cells, human stools, human biopsies, etc.), were based here on various percentages of tin, tungsten, titanium, niobium, tantalum and vanadium oxides. Sensor voltage responses were statistically analyzed also with the receiver operating characteristic (ROC) curves, that allowed the identification of the cut-off discriminating between healthy and tumor affected subjects for each sensor, leading to an estimate of sensitivity and specificity parameters. ROC analysis demonstrated that sensors employing tin and titanium oxides decorated with gold nanoparticles gave sensitivities up to 80% yet with a specificity of 70%.

An Array of MOX Sensors and ANNs to Assess Grated Parmigiano Reggiano Cheese Packs’ Compliance with CFPR Guidelines

Parmigiano Reggiano cheese is one of the most appreciated Italian foods on account of its high nutrient content and taste. Due to its high cost, these characteristics make this product subject to counterfeiting in different forms. In this study, an approach based on an array of gas sensors has been employed to assess if it was possible to distinguish different samples based on their aroma. Samples were characterized in terms of rind percentage, seasoning, and rind working process. From the responses of the sensors, five features were extracted and the capability of these parameters to recognize target classes was tested with statistical analysis. Hence, the performance of the sensors’ array was quantified using artificial neural networks. To simplify the problem, a hierarchical approach has been used: three steps of classification were performed, and in each step one parameter of the grated cheese was identified (firstly, seasoning; secondly, rind working process; finally, rind percentage). The accuracies ranged from 88.24% to 100%.

Fast Measurements with MOX Sensors: A Least-Squares Approach to Blind Deconvolution

Metal oxide (MOX) sensors are widely used for chemical sensing due to their low cost, miniaturization, low power consumption and durability. Yet, getting instantaneous measurements of fluctuating gas concentration in turbulent plumes is not possible due to their slow response time. In this paper, we show that the slow response of MOX sensors can be compensated by deconvolution, provided that an invertible, parametrized, sensor model is available. We consider a nonlinear, first-order dynamic model that is mathematically tractable for MOX identification and deconvolution. By transforming the sensor signal in the log-domain, the system becomes linear in the parameters and these can be estimated by the least-squares techniques. Moreover, we use the MOX diversity in a sensor array to avoid training with a supervised signal. The information provided by two (or more) sensors, exposed to the same flow but responding with different dynamics, is exploited to recover the ground truth signal (gas input). This approach is known as blind deconvolution. We demonstrate its efficiency on MOX sensors recorded in turbulent plumes. The reconstructed signal is similar to the one obtained with a fast photo-ionization detector (PID). The technique is thus relevant to track a fast-changing gas concentration with MOX sensors, resulting in a compensated response time comparable to that of a PID.

Real-Time Sensor Response Characteristics of 3 Commercial Metal Oxide Sensors for Detection of BTEX and Chlorinated Aliphatic Hydrocarbon Organic Vapors

The objective of this study was to examine the sensor response characteristics of three commercial Internet of Things (IoT) compatible metal oxide (MOx) sensors in preparation for the development of field-scale sensor networks for the real-time monitoring of volatile organic compounds (VOCs) in indoor environments located in proximity to brownfield sites. Currently, there is limited examination of such sensor responses to relevant mixtures of target VOCs, such as the common petrochemicals benzene, toluene, ethylbenzene, and xylene (BTEX), as well as chlorinated aliphatic hydrocarbon (CAH) contaminants such as tetrachloroethylene (PCE) and trichloroethylene (TCE) which are frequently associated with deterioration of indoor air quality. To address this, a study of three commercial metal oxide (MOx) sensors (SGP30, BME680, and CCS811) was undertaken to examine the sensor response characteristics of individual components as well as mixtures of each of the target BTEX and CAH chemicals over relevant indoor air concentrations within the operating range of the MOx sensors (0–6000 ppb). Our investigation revealed similar response patterns to those previously reported for the thick film MOx sensor to most individual target VOCs, however, response trends for mixtures were more difficult to discern. In general, the MOx sensors we examined demonstrated similar magnitude responses to the CAHs as BTEX compounds indicating reliable detection of CAHs.

MOX Sensors to Ensure Suitable Parameters of Grated Parmigiano Reggiano Cheese

Parmigiano Reggiano (PR) cheese is a long-ripened hard cheese made in Northern Italy registered as a Protected Designation of Origin (PDO) in the European Union. [...]

A Novel Modular System for Breath Analysis Using Temperature Modulated MOX Sensors

In this work, a new generation of gas sensing systems specially designed for breath analysis is presented. The developed system comprises a compact modular, low volume, temperature-controlled sensing chamber with three compartments that can host different sensor types. In the presented system, one compartment contains an array of 8 analog MOX sensors and the other two 10 digital MOX sensors each. Here, we test the system for the detection of low concentrations of several compounds.

Metal Oxide Sensors Applied to Industrial and Consumer Applications: Examples and Requirements for Successful Implementation

Metal Oxide (MOx) sensors have been the subject of intense research and development over theyears, and are widely used in industrial and commercial sensing applications [...]