An Electronic Equipment for Automatic Identification of Forest Seed Species

This work proposes an electronic equipment which identifies forest seeds for academic and research purposes. Existing integral solutions are prohibitively costly for silviculture laboratories used in forestry teaching. Thus, they must identify the seed by visual inspection, causing visual fatigue and results with low reliability. The state of the art proposes solutions using support vector machines, achieving a 98.82% accuracy for sunflower seeds. Other solutions extract morphological attributes of mussel seeds to identify up to 5 species with an accuracy of 95%. Most solutions only identify a single seed type with similar sizes. In this context, an electronic equipment is developed. It consists of an image acquisition enclosure, an electromechanical device to move a camera so different sizes of seeds can be imaged at different distances, and a single-board computer to control the image processing and artificial intelligence (convolutional neural network) algorithms. The equipment achieves an accuracy of 95%, which is satisfactory for potential users and silviculture specialists.