Detection of aphids in wheat fields using a computer vision technique

Aphids cause major damage in wheat fields resulting in significant yield losses. Monitoring aphid populations and the identification of aphid species provides important data related to pest population dynamics and integrated pest management. Manual identification and counting of wheat aphids is labour intensive, inefficient and subjective factors can influence its accuracy. A method of aphid identification and population monitoring based on digital images was developed. It used a maximally stable extremal region descriptor to simplify the background of field images containing aphids, and then used histograms of oriented gradient features and a support vector machine to develop an aphid identification model. This method was compared with five other commonly used methods of aphid detection; their performance was analysed using images with different aphid density, colour, or location on the plant. The results demonstrated that our new method provided mean identification and error rates of 86.81% and 8.91%, respectively, which is superior to other methods. The proposed method was easy-to-use and provides efficient and accurate aphid population data, and therefore can be used for aphid infestation surveys in wheat fields.