Pest inventory of a field is a way of knowing when the thresholds for pest control is reached. It is of increasing interest to use machine learning to automate this process, however, many challenges arise with detection of small insects both in traps and on plants. 

This thesis investigates the prospects of developing an automatic warning system for notifying a user of when certain pests are detected in a trap. For this, sliding window with histogram of oriented gradients based support vector machine were implemented. Trap detection with neural network models and a check size function were tested for narrowing the detections down to pests of a certain size. The results indicates that with further refinement and more training images this approach might hold potential for fungus gnat and rape beetles. 

Further, this thesis also investigates detection performance of Mask R-CNN and YOLOv5 on different insects in fields for the purpose of automating the data gathering process. The models showed promise for detection of rape beetles. YOLOv5 also showed promise as a multi-class detector of different insects, where sizes ranged from small rape beetles to larger bumblebees.