Towards Adulteration Detection in Medicinal Leaves Using an Effective Feature Extractor

Adulteration of therapeutic herbs is a modern evil that threatens the very foundations of Ayurveda. The endangered status of many medicinal plants and the focus on increasing profits by suppliers of Ayurvedic herbs result in adulteration. Medicinal leaves rich in therapeutic properties are being adulterated with other leaves of the same family with similar morphological characteristics. These leaves not only reduce the efficacy of drugs but can even be potentially harmful. This paper proposes an effective feature extractor SaHOG by cascading the Sato filter with Histogram of Oriented Gradients. Initially, the input images are converted to Lab colour space, and a contrast enhancement algorithm has been applied to the luminous channel. The merged three channels are given to SaHOG and a non-linear polynomial kernel support vector machine performs multiclass classification with an accuracy of 95.6%. The results were validated on another four benchmark leaf datasets, and the proposed approach improves the performance of leaf classification compared with the standard method. This has also been statistically proven by the Wilcoxon signed-rank test.