Identification of novel optical plant signals for the early detection of cannabis pests

点击翻译按钮获取中文摘要

Today, cannabis growers use visual inspection to identify pest and fungus threats. Those visual judgments also guide a grower's estimate of the potential economic risk and the appropriate treatment. Growers themselves emphasize that this approach is subjective, likely often inaccurate, and acknowledge that by the time the threat is visible to a human, the outbreak's economic risk is already substantial. The purpose of this research project is leveraging a collaboration between experts in entomology, agricultural economics, and visual recognition, to identify new optical "plant signals" based on spectral signatures that can be rapidly and inexpensively analyzed to enable early and accurate identification of threats to cannabis plants, as well as classify those threat in terms of potential economic damage. Visual recognition methods are increasingly being developed to classify plant health and threats. However, an important shortcoming of current approaches is that the algorithms are trained using visual signals that human-scouts are able to identify in the field. This prevents an algorithm from identifying new signals that may be invisible to a human. Moreover, virtually no research in this area has been conducted for cannabis since it only recently became legal in Canada. To address these shortcomings, we will use controlled experiments to expose cannabis plants to different levels of infestation and then train a machine learning algorithm to identify new signals for cannabis that reveal the infestation and its severity before it would otherwise be apparent to the human eye.