Scouting of Whiteflies in Tomato Greenhouse Environment Using Deep Learning

This study shows the possibilities of how to replace tedious human labor—scouting yellow sticky traps (YST) for whiteflies—using artificial cognitive vision, specifically deep convolutional network (CNN), as a part more complex system—BERABOT. The used CNN is Faster R-CNN trained by transfer learning substitute scouting when low whiteflies infection phase was targeted. training conducted on pictures taken inside heated and lighted tomato production greenhouse “Bezdínek Farm” in Dolni Lutyne, Czechia. Used were collected way planned future fully automated robotic applications BERABOT system. achieved results compared with professional phytopathologist. employee’s against phytopathologist accomplished root-mean-square error (RMSE) equal 4.23, while developed model evaluated be 5.83. presented here open up new frontiers further tuning leading potential substituting an employee(s) make less expensive labor dependent.