Effective plant discrimination based on the combination of local binary pattern operators and multiclass support vector machine methods

Accurate crop and weed discrimination plays a critical role in addressing the challenges of weed management in agriculture. The use of herbicides is currently the most common approach to weed control. However, herbicide resistant plants have long been recognised as a major concern due to the excessive use of herbicides. Effective weed detection techniques can reduce the cost of weed management and improve crop quality and yield. A computationally efficient and robust plant classification algorithm is developed and applied to the classification of three crops: Brassica napus (canola), Zea mays (maize/corn), and radish. The developed algorithm is based on the combination of Local Binary Pattern (LBP) operators, for the extraction of crop leaf textural features and Support vector machine (SVM) method, for multiclass plant classification. This paper presents the first investigation of the accuracy of the combined LBP algorithms, trained using a large dataset of canola, radish and barley leaf images captured by a testing facility under simulated field conditions. The dataset has four subclasses, background, canola, corn, and radish, with 24,000 images used for training and 6000 images, for validation. The dataset is referred herein as “bccr-segset” and published online. In each subclass, plant images are collected at four crop growth stages. Experimentally, the algorithm demonstrates plant classification accuracy as high as 91.85%, for the four classes. Keywords: Plant discrimination, Classification, LBP, PCA and SV