Non Technical Summary
Correct pest identification is crucial for successful implementation of integrated pest management in agriculture. Unfortunately, identification requires expert consultation and is time consuming. As a result, many pesticide applications are made unnecessarily, with associated environmental and economic costs, or fail to be made despite being justified. The goal of this project is to develop AI-based Decision Support System (DSS) for pests of the Inland Pacific Northwest small grain system. This DSS will allow farmers and other stakeholders to identify pests by taking a cellphone picture and uploading it to a mobile application.

Classification

Knowledge Area (KA)	Subject of Investigation (SOI)	Field of Science (FOS)	Percent
211	1549	1130	50%
211	1549	2080	50%

Knowledge Area
211 - Insects, Mites, and Other Arthropods Affecting Plants;

Subject Of Investigation
1549 - Wheat, general/other;

Field Of Science
1130 - Entomology and acarology; 2080 - Mathematics and computer sciences;

Goals / Objectives
The goalof this project is to createa website and mobile application that will allow identification and sharing of information on insect pests in Pacific Northwest cereal crop systems. More specifically, the objectives and sub-objectives of the project are:Objective 1. Create an open-source artificial intelligence software framework for automated identification of Inland PNW cereal system crop pests from cell phone photographs1a. Collect images of current and anticipated insect pests of Inland PNW cereal crops and rotational crops and train an artificial neural network (ANN) to classify them to species1b. Refine the image processing, ANN training, and prediction for efficiency in various visual contexts and pest combinations and to utilize real-time interactions with usersObjective 2. Incorporate the framework from (1) into an AI-aided decision support system (DSS) and community-based resource for managing pests in Inland PNW cereal systems.2a. Couple the identification framework with recommendations within a mobile application for use by producers and pest advisors2b. Build in the capacity for users to upload images into a web portal for community feedback and supervised inclusion in the ongoing training databaseObjective 3. Refine and disseminate the system developed in (2) to the target user populations3a. Solicit volunteer test users through extension outlets, grower meetings and other conduits3b. Release Version 1.0 and disseminate throughout the Inland PNW and in other regions with shared rotational crops and pest complexes

Project Methods
Objective 1:Collect images of crop pests from the Inland PNW region through field trips and collaborations.Utilize Python, Keras, and TensorFlow libraries for deep learning software development.Fine-tune a state-of-the-art neural network trained on the ImageNet dataset for pest recognition.Use data augmentation techniques to increase the training set and improve classifier performance.Objective 2:Develop a native cross-platform mobile application to allow users to upload images and receive real-time species classification.Create an interactive web application portal and a database for community feedback and inclusion of correctly identified images into the training database.Objective 3:Form an advisory board with growers, commodity commission representatives, extension educators, and industry representatives for feedback.Conduct in-person and online evaluations to test the effectiveness and accuracy of the identification tool and collect stakeholder feedback.Disseminate the optimized AI-based DSS to the Inland PNW stakeholders and other regions with shared rotational crops and pest complexes.