Small datasets for fruit detection with transfer learning

A common approach to the problem of fruit detectionin images is to design a deep learning network andtrain a model to locate objects, using bounding boxes to identifyregions containing fruit. However, this requires sufficient dataand presents challenges for small datasets. Transfer learning,which acquires knowledge from a source domain and brings thatto a new target domain, can produce improved performance inthe target domain. The work discussed in this paper shows theapplication of transfer learning for fruit detection with smalldatasets and presents analysis between the number of trainingimages in source and target domains. This investigation is basedon three datasets: two containing tomatoes and one containingstrawberries. Experimental results indicate that transfer learningcan enhance prediction with limited data.