Search

Recent works have shown that Convolutional Neural Networks (CNNs), because of their effectiveness in feature extraction and classification tasks, are suitable tools to address the Facial Expression Recognition (FER) problem. Further, it has been pointed out how ensembles of CNNs allow improving classification accuracy. Nevertheless, a detailed experimental analysis on how ensembles of CNNs could be effectively generated in the FER context has not been performed yet, although it would have considerable value for improving the results obtained in the FER task. This paper aims to present an extensive investigation on different aspects of the ensemble generation, focusing on the factors that influence the classification accuracy on the FER context. In particular, we evaluate several strategies for the ensemble generation, different aggregation schemes, and the dependence upon the number of base classifiers in the ensemble. The final objective is to provide some indications for building up effective ensembles of CNNs. Specifically, we observed that exploiting different sources of variability is crucial for the improvement of the overall accuracy. To this aim, pre-processing and pre-training procedures are able to provide a satisfactory variability across the base classifiers, while the use of different seeds does not appear as an effective solution. Bagging ensures a high ensemble gain, but the overall accuracy is limited by poor-performing base classifiers. The impact of increasing the ensemble size specifically depends on the adopted strategy, but also in the best case the performance gain obtained by involving additional base classifiers becomes not significant beyond a certain limit size, thus suggesting to avoid very large ensembles. Finally, the classic averaging voting proves to be an appropriate aggregation scheme, achieving accuracy values comparable to or slightly better than the other experimented operators.

Mapping the within-field variability of crop status is of great interest in precision agriculture that seeks to match agronomic inputs to crop demand, both spatially and temporally. In this context, nitrogen (N) management plays a key role that must balance its importance in crop production with its potential to be a source of environmental pollution. Remote and proximal sensing techniques are widely studied to assess the dynamics of crop status during the growing season. While many experiments were conducted to prove the feasibility of optical sensors to estimate N management-linked variables, a summary evaluation of their performance in maize is lacking. This review considers studies of ground-measured maize variables with remote and proximal optical sensor measurements under varying N levels to inform the feasibility of using sensors for N management. We collected 66 papers published between 1992 and May 2017 that reported 647 regressions between vegetation indices and maize variables (chlorophyll content, N concentration, leaf area index, above ground biomass, crop N uptake, crop yield, and optimum N rate). Regression tree analysis was applied to understand the roles of seven factors on the performance of estimation of maize variables with vegetation indices: crop development stage, sensor type, acquisition mode, sensed target, spatial resolution, type of vegetation index, and standardisation of sensor readings. Our results indicate that the factors have different effects accordingly to the different maize variables. Furthermore, regression parameters are specific for location, year, and cultivar. As empiricism severely limits practical application of these highly specific regressions, estimation of optimal N rates for delivery to the soil-crop system should take into account N budgets in soil that could be derived by chemical analysis or soil sensors as well as by more complex mechanistic simulation models.

Anaerobic digestion of animal manures, and separation of digestates into a liquid (LF) and a solid fraction (SF) affect manure N availability to crops. Digestates and LF potentially supply more readily available N to crops compared to undigested manures and SF, due to their higher ammonium content, lower C to organic N ratio and lower dry matter content. Studies on digestates decomposition in soil are still scarce compared to those on undigested manures. We therefore carried out an incubation experiment to measure CO2 emissions and soil mineral N dynamics following addition to soil of: ammonium sulphate (AS); digested cattle slurry-maize mix (DSMM); LF and SF of DSMM; undigested cattle slurry (US). The incubation lasted 181 days and was conducted at 25 °C and constant soil water content. Mineralization of C from US (51% of manure C after 181 days) was higher than that of digested manures (32–34%). Availability of applied ammonium after 181 days was similar for AS, DSMM and LF (70–78% of added ammonium), and was higher than that of SF and US (8 and 46% of added ammonium, respectively). Physical-chemical differences among manures explained main differences in decomposition dynamics. Indeed, US and SF induced net N immobilization (9–16% of manure organic N at day 181) due to high C to organic N ratio, high cellulose and volatile fatty acids content. Conversely, DSMM and LF induced net N mineralization (≈ 30% of manure organic N) due to a low C to organic N ratio and cellulose content. The findings of this laboratory experiment confirm the availability of manure N previously measured in a field experiment, where the same manures were applied to the same soil for the fertilization of silage-maize.

Accurate estimation of animal manure nitrogen (N) availability is required to maximize crop N use efficiency and reduce environmental N losses. Many field and laboratory experiments have shown that first-year net mineralization of manure organic N is often negligible, which often causes crop available N to approximate the ammonium N content of the manure. Anaerobic digestion increases the ammonium share and reduces the C to organic N ratio of animal manures, potentially increasing their N fertilizer value. In 2011, we undertook a three-year field experiment in Northern Italy to estimate the N fertilizer value of four manures: undigested cattle slurry, digested cattle slurry-maize mix, and liquid and solid fractions of the digested slurry-maize mix. The experiment also allowed us to test if ammonium recovery was similar among manures, and between manures and ammonium sulphate. Fertilizers were applied annually to plots before silage maize cultivation that was followed by an unfertilized Italian ryegrass crop. Results showed that the recovery of ammonium from manure in maize did not differ significantly compared to ammonium sulphate among all the fertilizers in 2013; however, in 2011 and 2012 it was significantly lower for all manures except digested slurry-maize mix and its liquid fraction in 2011. The increased recovery of applied N in 2012 and 2013 for solid fraction and undigested manure were likely due to the residual effect of previously applied organic N.

Classification is a functionality that plays a central role in the development of modern expert systems, across a wide variety of application fields: using accurate, efficient, and compact classification models is often a prime requirement. Boosting (and AdaBoost in particular) is a well-known technique to obtain robust classifiers from properly-learned weak classifiers, thus it is particularly attracting in many practical settings. Although the use of traditional classifiers as base learners in AdaBoost has already been widely studied, the adoption of fuzzy weak learners still requires further investigations. In this paper we describe FDT-Boost, a boosting approach shaped according to the SAMME-AdaBoost scheme, which leverages fuzzy binary decision trees as multi-class base classifiers. Such trees are kept compact by constraining their depth, without lowering the classification accuracy. The experimental evaluation of FDT-Boost has been carried out using a benchmark containing eighteen classification datasets. Comparing our approach with FURIA, one of the most popular fuzzy classifiers, with a fuzzy binary decision tree, and with a fuzzy multi-way decision tree, we show that FDT-Boost is accurate, getting to results that are statistically better than those achieved by the other approaches. Moreover, compared to a crisp SAMME-AdaBoost implementation, FDT-Boost shows similar performances, but the relative produced models are significantly less complex, thus opening up further exploitation chances also in memory-constrained systems.

Best management practices that could improve sustainability of dairy farming systems in northern Italy include crop rotation, green manure, sprinkler or drip irrigation, incorporation of crop residue, and adoption of a nutrient management plan. Despite the numerous advantages that scientific literature reports for these Best management practices, they are not always adopted by farmers, because other factors – of financial, technical, or social nature – limit their adoption. The theory of planned behaviour, based on the identification of outcomes, referents surrounding the farmers, and control factors, was applied through a detailed questionnaire to study individual farmer beliefs that influence the intention to adopt best practices. More than 50% out of the farms applied incorporation of crop residue, rotation with a grass or a legume meadow, sprinkler or drip irrigation, and adopted a nutrient management plan. Reasons for applying them were mainly related to soil sustainability (improvement of soil organic matter content, soil structure, fertility and yield) or to environmental sustainability (reduction of nitrogen losses, use of fertilizers, herbicides or insecticides). Among the main barriers to their adoption, the most important ones were an increase in direct or indirect costs. The only practice that was not adopted and, despite a limited number of barriers, will not be adopted by farmers, is green manure. Likely, our survey did not capture the real barriers against the adoption of this practice. Across all best management practices, the main difference between adopters and non-adopters was found in referents’ opinion on applying them. This means that it is very important, for the adoption of best management practices, that the community of family members, neighbor farmers, and various advisors, are in favour of adoption. This important finding should be used by public authorities to promote the development of focus groups, demonstration days, demonstration farms, and especially good and updated independent farm advisors who could substantially increase the adoption of best management practices by farmers.

Clay minerals can fix high amounts of ammonium in non-exchangeable form, which can be of management importance when some ammonium applied with fertilisers is sequestered and not available to the crop. Non-exchangeable ammonium can be released slowly and increase soil N availability. The percentage of ammonium fixed depends, among other factors, on the ammonium dose applied. We applied increasing amounts of ammonium to three soils and then measured (after 48 h) the soil equilibrium concentration of non-exchangeable, exchangeable, and soluble ammonium. At a 70 mg NH 4 –N kg − 1 dose, non-exchangeable ammonium measured 60, 24, and 17% on the three soils, while at 2800 mg NH 4 –N kg − 1 the fixation percentage decreased non-linearly to 12, 4, and 5%. Ammonium equilibrium concentrations in non-exchangeable, exchangeable, and soluble fractions as a function of ammonium applied were simulated with two innovative analytically-solved models using two Langmuir equations, one for each pair of pools. Both models predicted well the three ammonium fractions in all soils, with average relative root mean squared errors ranging from 2 to 5%. Sensitivity analysis demonstrated the importance of all four model parameters; however, to extend the model to other soils requires measurement of two parameters — cation exchange capacity and maximum fixation of ammonium in non-exchangeable form. The other two parameters are equilibrium constants that vary little among soils, according to our preliminary results. Therefore, to simplify model use, averaged optimised constant values on multiple soils might suffice. Additional experiments on more soils are needed to verify this hypothesis.

The degree of temporal coincidence in the development of winter wheat (Triticum aestivum L.) and the cereal leaf beetles Oulema melanopus (L.) and Oulema duftschmidi (Redtenbacher, 1874) was studied by means of explanatory phenology models. Temperature and photoperiod control crop development, whereas oviposition and development of eggs and larvae of the two beetles depend on temperature and crop phases. The models parametrized with literature data satisfactorily represented crop and prepupal insect development at several Swiss and Italian locations. The successfully validated models were used for representing multiannual crop and insect development at seven locations on a European transect between the Danube river in the North and the Po River in the South. Depending on temperature and photoperiod, the crop phases occurred at different time periods but were generally of similar durations. The shifting of the crop phases exposed the cereal leaf beetles to environmental conditions which were similar during oviposition and slightly different as the growing season progressed. The simulated oviposition and prepupal survivorship was much higher for O. melanopus than for O. duftschmidi but did not differ between the locations. The crop phase-dependent mortality (Mc) was consistently higher for O. duftschmidi than for O. melanopus, whose Mc increased with increasing altitude. The extent of coinciding development was investigated by means of the summed larval development rates divided by the summed wheat development rate. During the oviposition period the insect development was coincident with wheat development. With time progression, however, the temperature difference between the locations increased causing an incomplete coincidence in the development of wheat and cereal leaf beetles. These results support the hypothesis that the extent of coinciding development of the three species is largely controlled by temperature and photoperiodic conditions.

To improve the sustainability of agricultural systems of the Lombardia region (northern Italy), a mixed indicator-model-expert approach was used. Starting from the results of a previous assessment of current management (ACT) in dairy and arable farms, alternative management scenarios at field level were designed in order to reduce nitrogen (N) losses whilst maintaining or improving the environmental and economic sustainability at the farming system level. By working with a group of experts supported by a mechanisation model and a cropping system model, two alternative N management scenarios were defined following a step-by-step decision procedure. The first scenario (FERT) is an improvement of the current fertiliser management scheme, applied at the same crops as in ACT and aimed at maintaining the same yields. The second scenario (ROT) is based on changes in crop rotations by introducing new crops to reduce N losses and to maintain economic profitability. The sustainability of the two scenarios was assessed and compared with agro-ecological and economic indicators. The results of FERT, indicate that the application of adequate N management plans tuned to the production target and the promotion of best management practices may help to reduce N surplus and consequently to save fossil energy and to decrease the costs of production. In the ROT scenario, the introduction of alfalfa cultivation reduces N surplus on maize, whereas intensive double cropping systems (two crops harvested in 12 months) increase N surplus and require higher energy consumptions and production costs compared to cultivating a summer crop only. However, in rotational systems more favourable weed population dynamics are expected compared to ACT. Both alternative scenarios were not implemented in practice, but they are realistic and are consistent with results of experiments where management options similar to those introduced in FERT and ROT were tested. This work indicates that the rational integration between scientific tools (indicators and models) and expert knowledge is adequate to deal with complex farming and cropping systems, which require a multidisciplinary approach.

The analysis of post-vaccination serological data poses nontrivial issues to the epidemiologists and policy makers who want to assess the effects of immunisation programmes. This is especially true for infections on the path to elimination as is the case for measles. We address these problems by using Bayesian Normal mixture models fitted to antibody counts data. This methodology allows us to estimate the seroprevalence of measles by age and, in contrast to conventional methods based on fixed cut-off points, to also distinguish between groups of individuals with different degrees of immunisation. We applied our methodology to two serological samples collected in Tuscany (Italy) in 2003 and in 2005-2006 respectively, i.e., before and after a large vaccination campaign targeted to school-age children. Besides showing the impact of the campaign, we were able to accurately identify a large pocket of susceptible individuals aged about 13-14 in 2005-2006, and a larger group of weakly immune individuals aged about 20 in 2005-2006. These cohorts therefore represent possible targets for further interventions towards measles elimination.

Background Pertussis is an acute infectious illness, caused by the bacteria Bordetella pertussis and commonly known as “whooping cough”. Waning immunity after vaccination or after natural infection contributes significantly to the increasing incidence rates in adolescents and adults. Prevention of pertussis in industrialized countries is mainly based on immunization with acellular vaccines in combination with other antigens. A booster dose with an adult-formulation tetanus-diphtheria toxoid and acellular pertussis vaccine (Tdap) is now recommended for all adolescents by several countries, and replacement of the decennial Td dose with a single or more doses of Tdap is recommended for adults. Objective Our review aims at describing the current knowledge on the impact of acellular pertussis vaccination in adolescents and adults, with particular focus on specific risk groups: adolescents, pregnant women and their newborns, and health care workers (HCWs), and secondly at suggesting possible immunization strategies. Methods Data were retrieved by searches of Pubmed, references, from relevant articles and open-access websites. Results In countries where an adolescent booster dose was adopted, a certain decrease of incidence rates was observed. No serologic correlate of protection after immunization exists, but subjects with high antibody levels against pertussis antigens are less likely to develop the disease. Tdap vaccine was demonstrated to induce antibodies to pertussis antigens exceeding those associated with efficacy in infants, in both adolescents and adults. Tdap use in pregnant women seems to be safe and might represent a useful tool in order to prevent pertussis cases in the first months of life. Neonatal immunization with monovalent acellular pertussis vaccine can efficiently prime T and B cells and act as a basis for future immune responses. Cocooning strategies involving all those surrounding newborns have started to be implemented. Their impact on infant pertussis cases will be evaluated in the coming years. Coverage in HCWs should be increased, given their important role in pertussis transmission in health care settings. Conclusions Despite the more recent position paper of WHO gives priority to infant and childhood vaccination against pertussis and leaves adolescent, adult and risk group immunization as an option for the future, data are quickly accumulating to support the need to consider pertussis vaccination as a crucial preventative intervention even in adolescents and special risk groups.

To formulate best management practices for animal slurry, it is important to understand and predict its decomposition in the soil. Slurry decomposition dynamics can be studied by measuring CO 2 fluxes and soil mineral nitrogen concentration during laboratory incubations and subsequently calibrating a simulation model. Carbon and nitrogen dynamics are linked and both should be properly simulated. In this work we wanted to identify the tradeoffs between errors in the simulation of C respiration and of soil inorganic N concentration. We optimised six parameters of CN-SIM (a mechanistic dynamic simulation model), using data of respired C and soil inorganic N measured during a 180-day laboratory incubation of five dairy slurries on three soils. Optimisation was carried out with a multi-objective genetic algorithm (NSGA-II), by minimising the Relative Root Mean Squared Error (RRMSE) between observations and simulations. The simulation of C respiration was frequently conflicting with the simulation of inorganic N, i.e. low RRMSE–CO 2 corresponded with high RRMSE–N and vice versa. When minimising RRMSE–CO 2 a set of parameters was obtained that enhanced microbial N immobilisation and reduced the turnover of the organic pools, to match the observed decrease of inorganic N in the 28 days after slurry addition to soil. Remineralisation occurring in the following 150 days caused a marked overestimation of inorganic N. When minimising RRMSE–N, the optimisation provided parameters that strongly reduced remineralisation of immobilised N by markedly diminishing C respiration, with a consequent underestimation of CO 2 emission. A modified version of the model, containing a simple implementation of denitrification and of clay fixation/release of ammonium, performed better than the original model for most treatments. We conclude that the mineralisation/immobilisation turnover in the model is not fully adequate to represent C and N dynamics. We also discuss the implementation of changes (time-varying microbial efficiency and C to N ratio; simulation of ammonium clay fixation and emissions of N 2 /N 2 O) to improve model performance.

Upscaling of agro-ecological indicators applied in regional analyses is sensitive to scale issues of the input data. This study develops a methodology to quantify this sensitivity for an indicator of soil organic carbon (SOC) dynamics at the farming system level. A reference case consists of seven fully described farms in northern Italy. Both upscaling in complexity by substituting measured input with estimated input and upscaling in space by extending the methods to farms not included in the reference case are addressed. The indicator increased with 3–107% at four farms after substituting measured management input with that estimated by an expert, whereas it remained unchanged or decreased at the other three farms. Taking the modal value from a cluster of pedological input did not lead to additional uncertainty in most cases, and only slightly increased it in others. We evaluated spatial upscaling by including 733 farms divided in 18 clusters that were described with less information as compared to the reference farms. Within each cluster, we observed relevant variability of the indicator (coefficients of variation of 12–43%), as a consequence of the heterogeneity of farms comprised in each cluster. In each cluster we calculated the indicator for one virtual farm, defined by using modal values for basic farm inputs. In this case the indicator was highly correlated ( R 2 = 0.98) with the average of the values obtained using measured basic farm inputs. We conclude that upscaling in complexity and space introduces uncertainty in the values of the indicator compared to the reference case. The extent of such differences depends on the variability of the systems under analysis and on indicator sensitivity.

Objective: To assess the value of intraoperatory radioguided probe detection to guide surgical resection of malignant lesions previously detected by 18

Objective To assess the value of intraoperatory radioguided probe detection to guide surgical resection of malignant lesions previously detected by 18 F-FDG PET-CT. Material Twelve consecutive patients with suspected tumor recurrence detected by 18 F-FDG PET-CT considered resectable were enrolled in the study. Ultrasound guided fine needle aspiration (FNA) before surgery was performed in 6 patients and CT guided biopsy was performed in 1 patient. In 5 patients with accessible lesions, a radioguided occult lesion localization (ROLL) technique was performed after injection of 99m Tc-colloid (1.7–2.4 mCi) inside the lesion under ultrasound or CT guidance, pre-operatively. Radioguided surgical detection was then carried out 19–24 h afterwards using the gamma probe. In 7 patients with non-accessible needle lesions or multiple lesions, 9.5–10.5 mCi of 18 F-FDG were injected 3–5 h before radioguided surgery using a PET-dedicated probe (Gamma locator DXI-GF&E). Results ROLL technique: all lesions injected with nanocolloid were resected (6 lesions in 5 patients, 1 patient with 2 lesions), and recurrence was histologically confirmed. PET probe: fourteen out of 16 hypermetabolic lesions detected on the PET-CT were resected. One cervical and one mediastinal lymph node in different patients could not be excised. Histological recurrence was confirmed in 12 out of 14 lesions. In one patient, the 2 lymph nodes excised were inflammatory. Conclusions 18 F-FDG PET-CT can be key in deciding surgical approach and appropriate radioguided protocol. When lesions are solitary and easily accessible, ROLL technique seems the method of choice. PET probe is more adequate for less accessible lesions.

Scientific knowledge of carbon and nitrogen dynamics incorporated into soil organic matter (SOM) dynamic simulation models can be tested using experimental data from decomposition experiments. This possibility is particularly interesting for liquid animal manures, which are very heterogeneous materials. CN-SIM is a SOM model that was successfully applied for the short- and long-term simulation of C and N dynamics in the soil. The objective of this research was to test this model using results of a 180-d laboratory incubation of five liquid dairy manures in three soils. Initial sensitivity analysis indicated that model parameters that mostly impact net CO2 and soil mineral nitrogen concentration (SMN) are those that allocate manure C and N to model pools, their decomposition constants and soil parameters that describe microbial biomass and its residues. Automatic parameter optimisation was carried out with the downhill simplex method separately for 15 manure by soil combinations. After optimisation, CO2 respiration was well simulated for the entire period, while SMN was overestimated after day 40. Simulations overestimated the remineralisation of immobilised N in the first weeks. Simulated initial microbial immobilisation was the only possibility for the model to match the experimental decrease of SMN, while simulated N remineralisation after day 40 was necessary to enable a good simulation of respired CO2. Further work is needed to test whether these simulations can be improved by adding SMN sinks (such as denitrified N and clay-fixed ammonium).

In Lombardy region (northern Italy) agricultural systems are mostly based on cereals and forage crops, and are normally intensively cultivated. To monitor and analyse the most important agri-environmental issue of this area (water pollution by nitrates and plant protection products, non-renewable fossil energy exploitation, weed dynamics), we conducted an integrated agronomic, environmental and economic assessment of arable and dairy farming using indicators. The structural and management data about farming and cropping systems were collected by periodic interviews over a 3-year period in seven (three arable and four dairy) representative farms. The nitrogen surplus calculated at field scale ranged from low (27 kg N ha−1) to high (339 kg N ha−1) values, depending by the amount of chemical and organic fertilisers applied. Fossil inputs ranged from 11.2 to 46.0 GJ ha−1; the highest values were due to the high use of machinery and chemical inputs. The efficiency factor in energy transformation (an indicator of the dependence of food and feed production on non-renewable energy) ranged from 5.0 to 12.2. Large variability was also observed for economic performance (gross margin; from -364 to 1078 € ha−1). The lowest values were observed where the total costs of production (fixed and variable) were elevated. The simplification of cropping systems was responsible for a high probability of weed populations development. The values of Load Index, an indicator describing the potential toxicity of plant protection products on non-target organisms, showed an elevated use of active substances due to the dominance of maize. The indicator-based assessment, founded on a relatively simple data collection procedure, described crop management by combining different aspects into few quantities and highlighted its critical environmental issues. The work represents a starting point for improving cropping system management in the study area; the same procedure could be re-applied if data about improved management scenarios were available.

Resumen Objetivo Analizar los resultados conseguidos en diagnostico y estadificacion del cancer de pulmon (CP) por una unidad de diagnostico rapido de cancer de torax (UDR-CT) que incorpora la ultrasonografia endobronquial con puncion transbronquial aspirativa en tiempo real (USEB-PTBA-tr) a la evaluacion clinica del paciente previa al tratamiento. Metodo Se ha realizado un estudio observacional del conjunto de pacientes valorados por sospecha diagnostica de CP en una UDR-CT durante cuatro anos, registrando los tiempos y la tecnica requeridos para el diagnostico, el tiempo para el tratamiento, y el grado de identificacion de enfermedad en estadio inicial. Resultados Seiscientos setenta y ocho pacientes fueron atendidos en la UDR-CT, en 352 casos el diagnostico fue confirmado en una o mas muestras anatomopatologicas. En 170 pacientes el diagnostico se obtuvo con biopsias y/o citologias obtenidas por fibrobroncoscopia (48,2%), la USEB-TTBA-rt confirmo la sospecha clinica en 70 pacientes (19,9%). En 280 pacientes afectos de carcinoma de pulmon no celula pequena (CPNCP) se practicaron 166 USEB-PRBA-tr de estadificacion (59,3%) y en 105 de ellos la tecnica mostro unicamente enfermedad local (37,5%). En 83 de estos pacientes se procedio a cirugia terapeutica, que fue radical en 73 casos (87,9%). Conclusion En la mitad de pacientes remitidos a la UDR-CT por sospecha de CP el diagnostico se confirma, por tecnicas endoscopicas en tres cuartas partes de los casos. La USEB-TTBA-rt es la tecnica diagnostica en una quinta parte de los casos y de estadificacion en mas de la mitad de ellos, y permite reducir los tiempos de espera hasta el diagnostico y el inicio de tratamiento.