Agriculture has seen rapid development in both the quantity and quality of food production; however, the presence of pests and disease on crops can hamper the quality of agricultural pro-duce and can have a devastating effect on a farmer's bottom line. To combat the risk of pest infestations and disease, crop producers rely on a process known as crop scouting. Field-based crop scouting entails walking and surveying crop fields for yield reducing pests (insects, weeds, and diseases) and determining when control strategies must be taken. Typically, scouting is performed by either the farm operator or a contract service provided by local farm co-operatives or crop consulting businesses. The crop scouts hired by these contract service providers are often college interns trained through brief "scouting schools" and then sent into fields equipped with a paper-based scouting report and manuals to use for field pest identification. This traditional approach of crop scouting is cumbersome, time-consuming, inefficient, and prone to inaccurate pest identification. As smartphones and tablets are becoming more entrenched in the daily life of agricultural production, mobile-based scouting software apps have emerged. Though these apps provide a more user-friendly system to record and manage field data, pest identification remains inefficient, usually accomplished by answering dozens of questions. There is a critical need for auto-mated techniques to improve a user's scouting experience by making the path to identifying weeds, insects, or crop disorders easier, faster, and far more intuitive than at present. Allyke proposes VIRTUOSO (Visual Recognition of Pests for Crop Scouting), an image analysis technology for automatically identifying pests (insects, weeds, and diseases) during field-based crop scouting. VIRTUOSO accelerates the tedious, manual process of pest identification. Rather than scouring through large field pest identification manuals or answering dozens of often ambiguous questions posed by a mobile crop scouting app, VIRTUOSO helps crop scouts by automatically identifying the pest or crop disease from a photograph. VIRTUOSO uses ma-chine learning to learn a hierarchy of features that unveil salient feature patterns and hidden structure in the data. Each layer leads to progressively more abstract features at higher levels of the hierarchy. As a result, the learned representations are richer than existing handcrafted image features, making it easier to extract useful data when building classifiers or other predictors. Allyke has partnered with ScoutPro, a leading provider of mobile agricultural apps for crop scouting. ScoutPro has already established a critical mass of users for corn and soybean crop scouting applications. VIRTUOSO will be a cloud-based Software as a Service (SaaS) with a RESTful application programming interface (API). When integrated with ScoutPro's apps, we hypothesize that VIRTUOSO will not only improve scouting efficiency, but also the accuracy of the pest identification. It may also mean scouting happens more regularly, given the ease. Farmers will have a real-time understanding of crop pest pressures and gain the ability to make decisions more effectively, whether marketing their product, managing their risk, or just understanding the crop pest issues impacting crop production. The proposed innovation will not only represent a substantial breakthrough through the effective use of "Big Data" within agriculture and the USDA, but also other application domains where large image datasets are prevalent, including social media, retail, robotics, and medicine. It will enable the automatic derivation of analysis products that allow practitioners to quickly analyze data from new domains while greatly minimizing human effort.
