Akademik Veri Yönetim Sistemi
25th Signal Processing and Communications Applications Conference (SIU), Antalya, Türkiye, 15 - 18 Mayıs 2017
Insect infestations threaten yield efficiency in agricultural areas. Since insects massively reproduce, they not only reduce crop yield and quality, but expenditures made for biological pesticides form a huge portion of the total expenses. However, from the long-term perspective, blind chemical pest control on agricultural areas have turned out to be less than miraculous. Widespread adoption of chemical pesticides contributed to unprecedented increases in crop yields, but also resulted in the poisoning of farmworkers and rural residents, contamination of food and drinking water, destruction of wildlife habitats, and decimation of wildlife. Rather than chemical ones, using biotechnical approaches such as pheromone traps, a more effective and smarter pesticizing scenarios can be achieved if the reproduction stages of the insects can be observed. Using pheromone traps, the male insects are attracted to the trap. Hence massive reproduction is prevented, since males cannot mate with the female ones. However, pheromone traps require physical patrolling of the traps and the expensive labor cost due to this human labour is the most important disadvantage of the pheromone traps. Expert staff who can recognize different kinds of insects are required for the inspection of the traps. Many problems occur such as errors made in counting and recording of the collected data, because of the human factor in the whole cycle. To tackle with these problems, it is possible to integrate vision technology to the traps in order to assure more accurate record of the insect counts and types, as well as lower the labor costs. Hence, state of art computer vision techniques can be put into use for the automatic inspection of the visual data acquired through the traps. Our objective in this paper is to isolate and classify the insects in the traps under challenging environmental and illumination conditions using computer vision and machine learning algorithms. We first detect the insects, separate them from the background and extract the outer boundary of the insects. A variety of features are extracted and fused using weighted majority voting to obtain a decision for classification.