Having accurate, detailed, and up-to-date information about the location andbehavior of animals in the wild would revolutionize our ability to study andconserve ecosystems. We investigate the ability to automatically, accurately,and inexpensively collect such data, which could transform many fields ofbiology, ecology, and zoology into "big data" sciences. Motion sensor "cameratraps" enable collecting wildlife pictures inexpensively, unobtrusively, andfrequently. However, extracting information from these pictures remains anexpensive, time-consuming, manual task. We demonstrate that such informationcan be automatically extracted by deep learning, a cutting-edge type ofartificial intelligence. We train deep convolutional neural networks toidentify, count, and describe the behaviors of 48 species in the3.2-million-image Snapshot Serengeti dataset. Our deep neural networksautomatically identify animals with over 93.8% accuracy, and we expect thatnumber to improve rapidly in years to come. More importantly, if our systemclassifies only images it is confident about, our system can automate animalidentification for 99.3% of the data while still performing at the same 96.6%accuracy as that of crowdsourced teams of human volunteers, saving more than8.4 years (at 40 hours per week) of human labeling effort (i.e. over 17,000hours) on this 3.2-million-image dataset. Those efficiency gains immediatelyhighlight the importance of using deep neural networks to automate dataextraction from camera-trap images. Our results suggest that this technologycould enable the inexpensive, unobtrusive, high-volume, and even real-timecollection of a wealth of information about vast numbers of animals in thewild.
展开▼
