Forest regeneration: Individual tree crown detection techniques for density and stocking assessments

Sustainable forest management depends on successful forest regeneration. Th e use of remotely sensed aerial images or digitized aerial photographs of h igh spatial resolution could lead to accurate and timely semiautomatic comp uterized assessments. Techniques based on individual tree crown detection o r delineation can produce information about regenerating areas such as stem density, proper tree spacing and stocking, and even possibly, tree species and health estimations. Various computerized tree crown detection and delineation techniques alread y exist. Some are geared towards dense stands, while others are aimed at op en areas. An hybrid detection technique is able to detect the situation at hand and switch paradigm accordingly. Delineation techniques require higher spatial resolution and/or tree sizes, but offer more promises for tree spe cies recognition and health estimation. Most techniques can benefit from pa rticular acquisition conditions (e.g., autumn acquisition) and simple pre-p rocessing techniques to increase their detection or delineation capability and accuracy. This article describes two techniques presently under investigation by the authors: one of crown detection only, and another capable of crown delineat ion. Their strengths and weaknesses are illustrated and discussed, as are t heir pre-processing needs and image acquisition criteria. Various pre-proce ssing techniques are explored. Preliminary results with aerial images of re generation stands of various ages and densities demonstrate more quantitati vely these strengths and weaknesses relative to measurements made on the gr ound and from aerial photographs.