Sunn pest (Eurygaster integriceps put.) causes severe damage to wheat fields annually, reducing production by up to 50%. Rapid identification of pest concentration points and estimation of infestation levels in fields can be useful for production management and reducing the use of chemical sprays. Because of the limited ability to detect pests on the ground and access to high-resolution satellite imagery, aerial photography was considered for crop pest and disease detection. In this study, the feasibility of soft computing approaches and image processing to identify areas infected with sunn pest using near-infrared and visible light aerial imagery was investigated. An irrigated winter wheat field was surveyed for five consecutive months, from February to June. The spectral vegetation features (SVI), were extracted and analysed for both near infrared and visible light images. To detect infected spikes, Support Vector Machine (SVM) with Radial Basis Function (RBF) kernel was used. The Red and near-infrared (NIR) bands reflectance and, the Ratio Vegetation Index (RVI) for near-infrared images as well as Red and Green bands reflectance and normalised green blue difference index (NGBDI) for visible light images had the greatest impact on the performance of the SVM classifiers. The SVM classifiers were validated using the confusion matrix method. The best accuracy and performance of the detection system was achieved in February and March when the healthy wheat plant was still green. The mean accuracy for these two months was 0.97 and 0.93 for the SVM classifiers for NIR and visible light, respectively.

Keywords