array(3 items)
   publications => '25705' (5 chars)
   libraryUrl => '' (0 chars)
   layout => '0' (1 chars)

array(1 item)
   0 => Snowflake\Publications\Domain\Model\Publicationprototypepersistent entity (uid=25705, pid=124)
      originalId => protected25705 (integer)
      authors => protected'Oltramare, C.; Weiss, F. T.; Staudacher, P.; Kibirango,&
         nbsp;O.; Atuhaire, A.; Stamm, C.' (118 chars)
      title => protected'Pesticides monitoring in surface water of a subsistence agricultural catchme
         nt in Uganda using passive samplers' (111 chars)
      journal => protected'Environmental Science and Pollution Research' (44 chars)
      year => protected2023 (integer)
      volume => protected30 (integer)
      issue => protected'9' (1 chars)
      startpage => protected'10312' (5 chars)
      otherpage => protected'10328' (5 chars)
      categories => protected'smallholder farming; surface water; drinking water; pesticides; passive samp
         ling; environmental monitoring; Uganda; high-resolution mass spectrometry' (149 chars)
      description => protected'Convolutional Neural Networks (CNNs) offer an alternative to the image cross
         -correlation methods used in Particle Image Velocimetry (PIV) to reconstruct
          the fluid velocity field from the experimental recording. Despite the flexi
         bility of CNNs, the accuracy and robustness of the standard image processing
          remains unsurpassed for general PIV data. As CNNs are non-linear and typica
         lly entail up to millions of trainable parameters, they require large and ca
         refully designed training datasets to avoid over-fitting and to obtain resul
         ts that are accurate for a wide range of flow conditions and length scales. 
         Most training datasets consist of PIV-like data that are generated from disp
         lacement fields resulting from numerical flow simulations, which, in additio
         n of being computationally expensive, may be able to inform the network only
          about relatively few classes of flow problems. To overcome this issue and i
         mprove the accuracy of the velocity reconstructed by CNNs, we propose to tra
         in the networks with synthetic PIV-like data generated from random displacem
         ent fields. The underlying idea is that the training dataset simply needs to
          teach the network about the kinematic relationship between position and vel
         ocity. These kinematic training datasets are computationally inexpensive and
          may allow a much richer variability in terms of length scales by varying th
         e generation parameters. By training a state-of-the-art CNN, we investigate 
         the accuracy of the reconstructed displacement and velocity with synthetic a
         nd experimental test cases, such as a sinusoidal flow and wind-tunnel data f
         rom a turbulent-boundary-layer and a cylinder-wake experiment. We demonstrat
         e that kinematic training can drastically improve the accuracy of the CNN an
         d allows the network to outperform conventional cross-correlation methods, b
         eing more robust with respect to data noise and providing reconstructed velo
         city fields that have considerably higher spatial resolution (at pixel level
         ).' (1978 chars)
      serialnumber => protected'0944-1344' (9 chars)
      doi => protected'10.1007/s11356-022-22717-2' (26 chars)
      uid => protected25705 (integer)
      _localizedUid => protected25705 (integer)modified
      _languageUid => protectedNULL
      _versionedUid => protected25705 (integer)modified
      pid => protected124 (integer)