Weed detection using ultrasonic signal processing employing artificial neural network (ANN) with efficient extracted features

Abstract

In recent decades, different technologies such as image processing, spectral processing, and ultrasound techniques have been used to detect various weed species. In this paper, in addition to reviewing the conventional methods of weed detection, an alternative method based on processing of ultrasonic signals is introduced. In this regard, with the aid of a proper setup with the capability of sending and receiving 40 kHz ultrasonic waves, five weed species namely namely Portulacacea, Chenopodiumalbum L., Tribulusterrestris L.,  Amaranthusretroflexus L. and Salsolaiberica were identified .The continuous 40 kHz ultrasonic waves are sent to weed canopy and received back by an ultrasonic receiver. These signals are then transferred to a laptop (DELL INSPIRON 5010) and stored in MATLAB 2013a software for several signal features extraction, using artificial neural network (ANN) to discriminate the weeds and ultimately weed classification. Overall, the results showed that by eliminating about 20% of the inefficient signal features, the maximum detection accuracy of the ANN performance could be reached as high as 80%.