Monitoring Agricultural Lands in the Tropics Using Unsupervised Category-Specific Mesh Reconstruction and Deep Neural Networks
Academy for Mathematics, Science, and Engineering, United States
Monitoring tropical agricultural lands is important to manage sustainable food production, biodiversity and forestry. As the world’s population increases, we must have computational mechanisms to assess crops in agriculture. Recently, there have been developments within machine learning and computer vision, particularly in the 3D space. The approach of predicting 3D representation of objects in 2D imagery is crucial to many fields. In 2020, Goel et al. present Unsupervised category-specific mesh reconstruction, or U-CMR, which is an "analysis by synthesis" framework where the goal is to predict the likely shape, texture and camera viewpoint that could produce the image with various learned category-specific priors. Their particular contribution in this paper is a representation of the distribution over cameras, which we call "camera-multiplex". Instead of picking a point estimate, we maintain a set of camera hypotheses that are optimised during training to best explain the image given the current shape and texture. Separately, deep learning multi-temporal classification has become especially important in many interdisciplinary areas but specifically in crop classification. We train a convolutional neural network to classify images by crop category. In this work, we propose to combine these two methodologies, in other words to first classify and then create 3D representations using Unsupervised Category-Specific Mesh Reconstruction. This is an ongoing work and we hope to provide an enhanced novel computational way for farmers to gain insights into their crops. We hope that this work will lead to more productive food production and health outcomes in developing countries. Due to increased climate change around the world, many farmers in tropical and subtropical regions are experiencing the most devastating effect and must adapt to the changing time. Understanding trends in their crops through automated methods like the one we present here is a step in the right direction.
Keywords: Automation, computer vision, deep learning, image processing, machine learning
Contact Address: Thomas Chen, Academy for Mathematics, Science, and Engineering, 520 W Main St, 07866 Rockaway, United States, e-mail: thomaschen7acm.org