Marko Panic - Academia.edu (original) (raw)

Conference Presentations by Marko Panic

Spectral reflectance indices as a phenotyping tool for assessing morpho-physiological traits of winter wheat (Triticum aestivum L.), 2017

Spectral reflectance indices as a phenotyping tool for assessing morpho-physiological traits of w... more Spectral reflectance indices as a phenotyping tool for assessing morpho-physiological traits of winter wheat (Triticum aestivum L.)

Nataša Ljubičić1, Oskar Marko1, Ivana Maksimović2, Marko Panić1, Marina Putnik-Delić2, Marko Kostić2, Milena Daničić2, Sanja Brdar1, Radivoje Jevtić3 and Vladimir Crnojević1

1BioSense Institute, Dr Zorana Đinđića 1, 21 000 Novi Sad, Serbia
2 University of Novi Sad, Faculty of Agriculture, 21 000 Novi Sad, Serbia
3 Institute of Field and Vegetable Crops, 21 000 Novi Sad, Serbia

Morpho-physiological traits of wheat such as a grain weight per plant, total leaf chlorophyll content, carotenoids, relative dry matter and nitrogen content are important traits for the growth of winter wheat genotypes. However, methods to estimate these traits are laborious and destructive. Spectral reflectance indices based on combination of visible and near-infrared wavelengths such as NDVI (Normalized Difference Vegetation Index), represent one of the most promising tools for application in field phenotyping with potential to provide complex information on different morpho-physiological traits of wheat. The aim of this study was to assess the utility of NDVI measurements of wheat canopy in identification of a specific growth stage in which remotely sensed data show the largest correlation with final grain yield, grain weight per plant, total leaf chlorophyll and carotenoid content, relative dry matter and nitrogen content in 29 winter wheat (Triticum aestivum L.) genotypes. The NDVI was measured using an active hand-held sensor GreenSeeker (NTech Industries Inc., Ukiah, California, USA) and hyperspectral camera (Ximea Corp., Lakewood, CO USA) at four growth stages of wheat: full flowering (BBCH 65), medium milk (BBCH 75), early dough (BBCH 83) and fully ripe stage (BBCH 89). Overall 66 different hyperspectral NDVIs were calculated from two-band combinations between red (600-700 nm) or far red (700-740 nm) and near-infrared (756-946 nm) regions. Pearson’s correlation coefficient was used to explore the relationship among examined traits and NDVI measured at different growth stages of wheat. Obtained results indicate that most of observed NDVI indices showed negative correlation with the relative dry matter content at all observed growth stages. Significant positive correlations (higher than 0.6 and significant at P < 0.05) were found between the specific hyperspectral NDVIs measured at medium milk stage and grain weight per plant, total leaf chlorophyll, carotenoid and nitrogen content, as well as with final grain yield of wheat. The strong positive relationship between NDVI and examined traits found at medium milk stage suggests that this stage is the most appropriate for estimation of these traits of winter wheat in semiarid or similar wheat growing conditions. The overall results indicate that spectral reflectance tools based on combined visible and near-infrared wavelengths, such as NDVI, could be successfully applied to assess morpho-physiological traits of a large number of winter wheat genotypes in a rapid and non-destructive manner. Furthermore, although neither device appeared to have a sizeable advantage over the other, NDVI acquired by hyperspectral camera does appear to be more indicative than NDVI acquired by GreenSeeker sensor, suggesting that alternative spectral combinations can be used in assessing targeted traits of winter wheat genotypes.

Morpho-physiological traits of wheat such as a grain weight per plant, total leaf chlorophyll con... more Morpho-physiological traits of wheat such as a grain weight per plant, total leaf chlorophyll content, carotenoids, relative dry matter and nitrogen content are important traits for the growth of winter wheat genotypes. However, methods to estimate these traits are laborious and destructive. Spectral reflectance indices based on combination of visible and near-infrared wavelengths such as NDVI (Normalized Difference Vegetation Index), represent one of the most promising tools for application in field phenotyping with potential to provide complex information on different morpho-physiological traits of wheat. The aim of this study was to assess the utility of NDVI measurements of wheat canopy in identification of a specific growth stage in which remotely sensed data show the largest correlation with final grain yield, grain weight per plant, total leaf chlorophyll and carotenoid content, relative dry matter and nitrogen content in 29 winter wheat (Triticum aestivum L.) genotypes. The NDVI was measured using an active hand-held sensor GreenSeeker (NTech Industries Inc., Ukiah, California, USA) and hyperspectral camera (Ximea Corp., Lakewood, CO USA) at four growth stages of wheat: full flowering (BBCH 65), medium milk (BBCH 75), early dough (BBCH 83) and fully ripe stage (BBCH 89). Overall 66 different hyperspectral NDVIs were calculated from two-band combinations between red (600-700 nm) or far red (700-740 nm) and near-infrared (756-946 nm) regions. Pearson’s correlation coefficient was used to explore the relationship among examined traits and NDVI measured at different growth stages of wheat. Obtained results indicate that most of observed NDVI indices showed negative correlation with the relative dry matter content at all observed growth stages. Significant positive correlations (higher than 0.6 and significant at P < 0.05) were found between the specific hyperspectral NDVIs measured at medium milk stage and grain weight per plant, total leaf chlorophyll, carotenoid and nitrogen content, as well as with final grain yield of wheat. The strong positive relationship between NDVI and examined traits found at medium milk stage suggests that this stage is the most appropriate for estimation of these traits of winter wheat in semiarid or similar wheat growing conditions. The overall results indicate that spectral reflectance tools based on combined visible and near-infrared wavelengths, such as NDVI, could be successfully applied to assess morpho-physiological traits of a large number of winter wheat genotypes in a rapid and non-destructive manner. Furthermore, although neither device appeared to have a sizeable advantage over the other, NDVI acquired by hyperspectral camera does appear to be more indicative than NDVI acquired by GreenSeeker sensor, suggesting that alternative spectral combinations can be used in assessing targeted traits of winter wheat genotypes.

Papers by Marko Panic

Applied Artificial Intelligence, Dec 21, 2022

Zenodo (CERN European Organization for Nuclear Research), Jun 16, 2022

Remote Sensing, Dec 14, 2022

This article is an open access article distributed under the terms and conditions of the Creative... more This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY

Science of The Total Environment, Jun 1, 2022

This is the first time that atmospheric concentrations of individual pollen types have been recor... more This is the first time that atmospheric concentrations of individual pollen types have been recorded by an automatic sampler with 1-hour and sub-hourly resolution (i.e. 1-minute and 1-second data). The data were collected by traditional Hirst type methods and state-of the art Rapid-E real-time bioaerosol detector. Airborne pollen data from 7 taxa, i.e. Acer negundo, Ambrosia, Broussonetia papyrifera, Cupressales (Taxaceae and Cupressaceae families), Platanus, Salix and Ulmus, were collected during the 2019 pollen season in Novi Sad, Serbia. Pollen data with daily, hourly and sub-hourly temporal resolution were analysed in terms of their temporal variability. The impact of turbulence kinetic energy (TKE) on pollen cloud homogeneity was investigated. Variations in Seasonal Pollen Integrals produced by Hirst and Rapid-E show that scaling factors are required to make data comparable. Daily average and hourly measurements recorded by the Rapid-E and Hirst were highly correlated and so examining Rapid-E measurements with sub-hourly resolution is assumed meaningful from the perspective of identification accuracy. Sub-hourly data provided an insight into the heterogenous nature of pollen in the air, with distinct peaks lasting ~5-10 min, and mostly single pollen grains recorded per second. Short term variations in 1-minute pollen concentrations could not be wholly explained by TKE. The new generation of automatic devices has the potential to increase our understanding of the distribution of bioaerosols in the air, provide insights into biological processes such as pollen release and dispersal mechanisms, and have the potential for us to conduct investigations into dose-response relationships and personal exposure to aeroallergens.

Scientific Reports, Feb 24, 2023

Pollen monitoring have become data-intensive in recent years as real-time detectors are deployed ... more Pollen monitoring have become data-intensive in recent years as real-time detectors are deployed to classify airborne pollen grains. Machine learning models with a focus on deep learning, have an essential role in the pollen classification task. Within this study we developed an explainable framework to unveil a deep learning model for pollen classification. Model works on data coming from single particle detector (Rapid-E) that records for each particle optical fingerprint with scattered light and laser induced fluorescence. Morphological properties of a particle are sensed with the light scattering process, while chemical properties are encoded with fluorescence spectrum and fluorescence lifetime induced by high-resolution laser. By utilizing these three data modalities, scattering, spectrum, and lifetime, deep learning-based models with millions of parameters are learned to distinguish different pollen classes, but a proper understanding of such a black-box model decisions demands additional methods to employ. Our study provides the first results of applied explainable artificial intelligence (xAI) methodology on the pollen classification model. Extracted knowledge on the important features that attribute to the predicting particular pollen classes is further examined from the perspective of domain knowledge and compared to available reference data on pollen sizes, shape, and laboratory spectrofluorometer measurements. In Europe as much as 40 percentage of population is affected by pollen allergy 1. The substantial costs from the disease itself or from productivity loss due to poor management of the disease exceeds several tens of billions euros per year 2. The burden of allergic disease can be limited by avoiding allergen exposure or timely therapy, which makes airborne pollen data and forecasts of utmost value both for patients and medical workers. Detection and quantification of airborne pollen have mainly been carried using standard volumetric method (EN16868) 3 which relies on labour intensive and lengthy manual identification of each bioaerosol particle under microscope resulting in at least 36 h delays for data availability. The stakeholders showed the need for the near real-time data 4 since it is expected to help patients relate better their symptoms to exposure thus providing a tool for more accurate timely diagnosis and for better assessment of therapy efficiency. In addition, like in meteorology, near real-time observations can be integrated into numerical models to provide improved spatial forecasts. Recent technological developments proved that sampling and characterizing single bioaerosol particles is possible 5,6 , however the discrimination is still challenging especially when pollen identification relies on complex signals representing both morphology and chemical composition of detected particles. The first attempt to resolve pollen classes from optical pollen monitoring based on time-resolved scattering and fluorescence was performed with artificial neural network and support vector machines classifiers 7. This classical machine learning approach demanded for extensive feature engineering steps for extracting properties of the measured signals. Further development of pollen classification models from chemical signatures and scattering information was accomplished with deep learning approach based on convolutional neural network (CNN) architecture 8 .

Scientific Reports, Nov 30, 2021

Tomato is an important commercial product which is perishable by nature and highly susceptible to... more Tomato is an important commercial product which is perishable by nature and highly susceptible to fungal incidence once it is harvested. Not all tomatoes are equally vulnerable to pathogenic fungi, and an early detection of the vulnerable ones can help in taking timely preventive actions, ranging from isolating tomato batches to adjusting storage conditions, but also in making right business decisions like dynamic pricing based on quality or better shelf life estimate. More importantly, early detection of vulnerable produce can help in taking timely actions to minimize potential post-harvest losses. This paper investigates Near-infrared (NIR) hyperspectral imaging (1000-1700 nm) and machine learning to build models to automatically predict the susceptibility of sepals of recently harvested tomatoes to future fungal infections. Hyperspectral images of newly harvested tomatoes (cultivar Brioso) from 5 different growers were acquired before the onset of any visible fungal infection. After imaging, the tomatoes were placed under controlled conditions suited for fungal germination and growth for a 4-day period, and then imaged using normal color cameras. All sepals in the color images were ranked for fungal severity using crowdsourcing, and the final severity of each sepal was fused using principal component analysis. A novel hyperspectral data processing pipeline is presented which was used to automatically segment the tomato sepals from spectral images with multiple tomatoes connected via a truss. The key modelling question addressed in this research is whether there is a correlation between the hyperspectral data captured at harvest and the fungal infection observed 4 days later. Using 10-fold and group k-fold cross-validation, XG-Boost and Random Forest based regression models were trained on the features derived from the hyperspectral data corresponding to each sepal in the training set and tested on hold out test set. The best model found a Pearson correlation of 0.837, showing that there is strong linear correlation between the NIR spectra and the future fungal severity of the sepal. The sepal specific predictions were aggregated to predict the susceptibility of individual tomatoes, and a correlation of 0.92 was found. Besides modelling, focus is also on model interpretation, particularly to understand which spectral features are most relevant to model prediction. Two approaches to model interpretation were explored, feature importance and SHAP (SHapley Additive exPlanations), resulting in similar conclusions that the NIR range between 1390-1420 nm contributes most to the model's final decision. Tomato is a popular and commercially important horticultural produce worldwide 1. Quality of tomato depends on growing conditions and chain conditions like humidity and temperature, as well as crop handling during harvest and post-harvest processes (transport, packaging, storage, processing etc.) 2. Like many other perishable fruits and vegetables, it is highly prone to post-harvest losses, reaching up to 30% in some developing countries 3. Early detection of disease has the potential to prevent losses because early actions can be taken to limit bigger damages (see e.g. 4). Tomato is known to be highly susceptible to pathogenic fungi, such as Penicillium, Aspergillus and Mucor, which tend to attack crops with high moisture and nutrient content 5-8. The weakening and damage to tomato tissue can be caused by specific environmental conditions (humidity and temperature) as well as due to poor product handling. This creates a potential entrance for fungal spores which, given appropriate germination conditions, may infect the stem, calyx, sepals, or tomato skin.

Zenodo (CERN European Organization for Nuclear Research), Nov 26, 2021

Zenodo (CERN European Organization for Nuclear Research), Sep 14, 2022

Glaciers are not only important indicators of the ongoing climate change, but are also highly rel... more Glaciers are not only important indicators of the ongoing climate change, but are also highly relevant as water storages and in connection with natural hazards. Nevertheless, mid-or long-term field monitoring is only conducted on rather few glaciers, which are mainly concentrated in a few areas on the globe. Field work in remote mountain areas often time consuming and expensive but reaching remote glaciers, e.g. in the Himalayas or in Antarctica, can also be dangerous or even impossible. Therefore remote sensing has been a main instrument in glacier monitoring for more than a decade. In this study, we investigate the use of multi-temporal high (Sentinel-2) and very high (Pléiades) resolution optical satellite data for the monitoring of glacier surfaces and their changes in the Khumbu-Himal (Nepal), the Ötztal Alps (Austria and Italy) and the Ortler-Cevedale Group (Italy). We classify the surfaces of different glaciers into classes relevant for mass balance calculations, which are snow, firn, ice, debris and water and try to detect changes in those surface classes over time.

Zenodo (CERN European Organization for Nuclear Research), Mar 16, 2017

Springer Optimization and Its Applications

Zenodo (CERN European Organization for Nuclear Research), Nov 26, 2021

The world's growing population is putting an immense pressure on agriculture to produce more with... more The world's growing population is putting an immense pressure on agriculture to produce more with less. In the context of conflicting economic, environmental and societal demands, decision-making across the whole supply chain needs to be optimised. In order to make informed decisions, data coming from satellites, drones, sensors and other sources needs to be analysed. However, due to complexity and magnitude of data, advanced machine learning and data analytics algorithms need to be employed. OBJECTIVES: This paper tackles two critical tasks in precision agriculture-management zone delineation and yield prediction. Management zones are regions in the field that have large inter-region and small intra-region variability, meaning that their boundaries divide the field into homogeneous zones for which the agronomic operations should be separately adjusted. Secondly, yield prediction is essential for fertiliser optimisation and post-harvest logistics. Fertiliser type and amount are tied to the amount of nutrients extracted from the soil and in order to compensate for this, nutrients need to be replenished. The information about the yield is also key for optimising harvesting, logistics, storage and sales. METHOD / DESIGN: The choice of input data depends on the use-case, but generally, there is a trade-off between precision and scalability. Within the scope of image processing, drones may provide high-resolution data, but their use is limited by the need of physical presence of the human operator in regular intervals during the season. Setninel-2 satellites on the other hand provide images at a 10 m resolution, but cover the whole globe every 5 days on average. For this reason, we chose them as the input data source. For management zone delineation, we calculated different spectral vegetation indices from satellite images, and applied the k-means algorithm. The resulting maps were post-processed so that the resolution of the zones fits the width of the fertiliser/pesticide spreader. Yield prediction was set on a per pixel basis. We used the soya yield maps from combine harvesters acquired in the years 2018-2020 for model training (411 ha in total) and a number of machine learning models were implemented, such as random forest, artificial neural networks, XGBoost and stochastic gradient descent. RESULTS: Accuracy of yield prediction algorithms was validated on the test set which included 14 out of 142 soya fields from the database. With the Pearson correlation coefficient of 0.74 and mean absolute error of 0.49 t/ha, stochastic gradient descent achieved the best performance. As for management zone delineation, the tool cannot be validated on a similar basis, as there is no objective division of the field into zones. Rather than that, we left the algorithm parameters, such as choice of the spectral index (from a number of soil and plant-based indices), the number of zones and the width of the machine, to the user to decide on, according to his/her preference, experience and the desired output. CONCLUSIONS: The aforementioned machine learning models are essential tools for monitoring crop growth. The resulting maps provide precious information to the farmers, who can optimise their decisions based on them. In order to facilitate rapid transfer of technology from academia to industry, we implemented a management zone delineation module within AgroSense. With more than 20,000 users, and ¼ of all Serbian farmland managed through the system, this technology transfer signifies an important step in digital transformation of agriculture.

With an increasing interest of the agricultural community in precision agriculture, this paper ai... more With an increasing interest of the agricultural community in precision agriculture, this paper aims to compare two novel sensing approaches for crop monitoring. The recently developed multispectral proximal sensor named Plant-O-Meter and Sentinel-2 satellite, which carries a multispectral optical instrument, are two sensors suitable for agricultural applications. Each of them has pros and cons regarding spatial, spectral and temporal resolutions and their complementary use will surely bring added value compared to information retrieved by a single sensor. In order to correctly address the problem of data fusion, compatibility studies between the two sensors are necessary. In this study, a maize field was sensed on several dates in 2018 growing season using both sensors. Numerous vegetation indices based on different spectral channel combinations were calculated and the results were compared using linear regression analysis. First results showed good positive correlations between the...

Aerobiologia, 2018

Forecasting daily airborne pollen concentrations is of great importance for management of seasona... more Forecasting daily airborne pollen concentrations is of great importance for management of seasonal allergies. This paper explores the performance of the pollen calendar as the most basic observation-oriented model for predicting daily concentrations of airborne Ambrosia, Betula and Poaceae pollen. Pollen calendars were calculated as the mean or median value of pollen concentrations on the same date in previous years of the available historic dataset, as well as the mean or median value of pollen concentrations of the smoothed dataset, pre-processed using moving mean and moving median. The performance of the models was evaluated by comparing forecasted to measured pollen concentrations at both daily and 10-day-average resolutions. This research demonstrates that the interpolation of missing data and pre-processing of the calibration dataset yields lower prediction errors. The increase in the number of calibration years corresponds to an improvement in the performance of the calendars in predicting daily pollen concentrations. However, the most significant improvement was obtained using four calibration years. The calendar models correspond well to the shape of the pollen curve. It was also found that daily resolution instead of 10-day averages adds to their value by emphasising variability in pollen exposure, which is important for personal assessment of doseresponse for pollen-sensitive individuals.

FlexiGroBots - Blueberry orchard UAV dataset Acquisition date: 02.07.2021.<br> Location: Ba... more FlexiGroBots - Blueberry orchard UAV dataset Acquisition date: 02.07.2021.<br> Location: Babe, Serbia Dataset consists of UAV drone images:<br> - 6 channels: reflectance blue, reflectance green, reflectance red, reflectance red edge, reflectance NIR, RGB. In order to align different channels, registration was applied and because of uneven illumination during the acquisition process, illumination correction was performed. Thus, the results of each preprocessing step are located in separate folders, i.e. raw data are in 100FPLAN and 101FPLAN, results of registration are in 100FPLAN_registrated and 101FPLAN_registrated, while images with corrected illumination are in 100FPLAN_registrated_corrected and 101FPLAN_registrated_corrected. Orthomosaics created before and after preprocessing are located in UAV orthomosaics folder.

Sensors

This paper presents an autonomous robotic system, an unmanned ground vehicle (UGV), for in-field ... more This paper presents an autonomous robotic system, an unmanned ground vehicle (UGV), for in-field soil sampling and analysis of nitrates. Compared to standard methods of soil analysis it has several advantages: each sample is individually analyzed compared to average sample analysis in standard methods; each sample is georeferenced, providing a map for precision base fertilizing; the process is fully autonomous; samples are analyzed in real-time, approximately 30 min per sample; and lightweight for less soil compaction. The robotic system has several modules: commercial robotic platform, anchoring module, sampling module, sample preparation module, sample analysis module, and communication module. The system is augmented with an in-house developed cloud-based platform. This platform uses satellite images, and an artificial intelligence (AI) proprietary algorithm to divide the target field into representative zones for sampling, thus, reducing and optimizing the number and locations o...

Spectral reflectance indices as a phenotyping tool for assessing morpho-physiological traits of winter wheat (Triticum aestivum L.), 2017

Nataša Ljubičić1, Oskar Marko1, Ivana Maksimović2, Marko Panić1, Marina Putnik-Delić2, Marko Kostić2, Milena Daničić2, Sanja Brdar1, Radivoje Jevtić3 and Vladimir Crnojević1

Applied Artificial Intelligence, Dec 21, 2022

Zenodo (CERN European Organization for Nuclear Research), Jun 16, 2022

Remote Sensing, Dec 14, 2022

Science of The Total Environment, Jun 1, 2022

Scientific Reports, Feb 24, 2023

Scientific Reports, Nov 30, 2021

Zenodo (CERN European Organization for Nuclear Research), Nov 26, 2021

Zenodo (CERN European Organization for Nuclear Research), Sep 14, 2022

Zenodo (CERN European Organization for Nuclear Research), Mar 16, 2017

Springer Optimization and Its Applications

Zenodo (CERN European Organization for Nuclear Research), Nov 26, 2021

Aerobiologia, 2018

Sensors

Remote Sensing

Agriculture is the backbone and the main sector of the industry for many countries in the world. ... more Agriculture is the backbone and the main sector of the industry for many countries in the world. Assessing crop yields is key to optimising on-field decisions and defining sustainable agricultural strategies. Remote sensing applications have greatly enhanced our ability to monitor and manage farming operation. The main objective of this research was to evaluate machine learning system for within-field soya yield prediction trained on Sentinel-2 multispectral images and soil parameters. Multispectral images used in the study came from ESA’s Sentinel-2 satellites. A total of 3 cloud-free Sentinel-2 multispectral images per year from specific periods of vegetation were used to obtain the time-series necessary for crop yield prediction. Yield monitor data were collected in three crop seasons (2018, 2019 and 2020) from a number of farms located in Upper Austria. The ground-truth database consisted of information about the location of the fields and crop yield monitor data on 411 ha of fa...

Lecture notes in networks and systems, Nov 19, 2022