Identification of weed seeds species in mixed sample with wheat grains using SIFT algorithm

Abstract

The problem of plant seed identification is important for agricultural sector, such as maintaining seed quality and to prevent the spreading of weed species. Seed identification is currently performed by a human seed analyst; human must often search through many seed images before finding the desired seed. This process of manual identification is slow and posses a degree of subjectivity which is hard to be quantified. Therefore, it is highly recommended economically to introduce an automatic system for seed identification. Modern techniques in different computer science fields such as image analysis, pattern recognition and computer vision can be applied in this system. In this paper, we use Scale-Invariant Feature Transform (SIFT) algorithm to identification three types of weed seeds (Coronopus didymus (L.) Sm., Lolium multiflorum Lam. and Chenopodium ambrosioides L.) that mixed with wheat grains samples. The accuracies of weed seeds detection were 90.5%, 89.2 and 95.3 for the three species respectively. SIFT algorithm discriminated well between wheat grains and weed seeds.