Image Processing Performance Assessment Using Crop Weed Competition Models

Abstract.

Precision treatment of both crops and weeds requires the accurate identification of both types of plant. However both identification and treatment methods are subject to error and it is important to understand how misclassification errors affect crop yield. This paper describes the use of a conductance growth model to quantify the effect of misclassification errors caused by an image analysis system.

Colour, morphology and knowledge about planting patterns have been combined, in an image analysis algorithm, to distinguish crop plants from weeds. As the crop growth stage advances, the algorithm is forced to trade improved crop recognition for reduced weed classification. Depending on the chosen method of weed removal, misclassification may result in inadvertent damage to the crop or even complete removal of crop plants and subsequent loss of yield. However incomplete removal of weeds might result in competition and subsequent yield reduction. The plant competition model allows prediction of final crop yield after weed or crop removal. The competition model also allows the investigation of the impact on yield of misclassification in the presence of both aggressive and benign weed types. The competition model and the image analysis algorithm have been linked successfully to investigate a range of misclassification scenarios in scenes containing cabbage plants.