Local feature-based identification and classification for orchard insects

Insect monitoring in orchards under integrated pest management mainly relies on traps and manual evaluation. Automation, including machine vision combined with pattern recognition has achieved some applications in areas such as fruit sorting, robotic harvesting and quality detection, etc. An invariant local feature-based insect classification method has been proposed to automatically classify certain common insects in orchards. An invariant region feature detector was used to extract local features and a scale invariant feature transform (SIFT) descriptor was adapted to represent features obtained from the detector. To represent the whole image, the bag of words method was introduced to cluster and form a visual word expression for each of the image objects, which is normalized to feature vectors as input of the classifiers. Samples of five common pest species in orchards, Cydia pomonella, Choristoneura rosaceana, Platynota idaeusalis, Argyrotaenia velutinana, Grapholita prunivora were used to verify the classification method. Performances of six classifiers, which were minimum least square linear classifier (MLSLC), K nearest neighbour classifier (KNNC), Parzen density based linear classifier (PDLC), principal component analysis expansion linear classifier (PCALC), nearest mean classifier (NMC), and support vector machine (SVM) were compared by classification result. The best classification results under 10-fold cross-validation test were 4.57% and 5.95% using PCALC and SVM, indicating that the local region detector based insect classification method could be an effective way for insect identification and classification.