Case Study: Diagnosing Crop Disease with Machine Learning

Case Study

The volatile combination of a rapidly growing population with an increasingly variable climate poses a serious threat to the agriculture industry. These complex challenges require innovative solutions that enable efficient production of food in environmentally responsible, yet practical and economical, ways.

Relevant Background

Among the greatest challenges farmers face are those conditions that are outside of their control, such as weather, soil quality, weeds, and crop disease. These factors inhibit growth of both biofuel crops and healthy produce for consumption.

Losses due to crop disease, whether caused by fungi, bacteria, or viruses, can be disastrous to the grower and the economy, and they can threaten the global population’s access to nutrition.

Crop disease can devastate natural ecosystems, leading to environmental problems, such as habitat loss for certain species. Additionally, in extreme cases, diseased crops can produce toxins that result in serious health problems for consumers.

The Challenge 

While farmers spend billions of dollars on crop-disease management, they often do so without access to adequate diagnostic tools. This can leave them using one-size-fits-all solutions that not only fail to address their specific needs, but often result in undesirable side effects, such as pollution and reduction of beneficial pollinator species, such as bees, butterflies, and others.

Our client knew that accurate diagnosis was the best way to target the exact solutions necessary to maintain crop health. However, without deep subject matter expertise, crop disease is difficult ⁠— nearly impossible ⁠— to diagnose. 

Without a physical examination of each specimen, our client was unable to recommend customized treatment options to customers. There was no way for plant pathologists to travel to every farm in need of assistance; the only solution was to bring the data to the pathologists.


We assisted with the architecture and development of a mobile application that enables farmers to photograph diseased crops and receive accurate diagnostics in near real time.

The diagnostic process occurs through the use of advanced image analytics powered by machine learning.

Solution Overview

Machine Learning: Disease classification capabilities were implemented using the advanced deep learning capabilities of Google's ML Engine. We used Google's high-performance platform and more than 50,000 images to train the neural network. With Google's Tensorflow Processing Units (TPUs), the training time of the neural network is greatly reduced, allowing for rapid and cost-efficient model updates as additional images are collected and curated.

Mobile Application: The mobile application allows a farmer to take photographs of the diseased leaves of their crops. These images are passed to the GCP-hosted ML services, and a diagnosis is returned to the farmer immediately. As a hosted, serverless platform, ML Engine provides scalability without the need to manage a set of servers.

Data Engineering: A reusable platform enables data and model management, along with DevOps support for multi-cloud infrastructure. This infrastructure assists data scientists in performing data cleansing, learning, and service deployment of solutions, utilizing AI analytics techniques at scale. The solution leverages GCP Cloud Storage for managing plant images and generated models, along with GCP Data Flow for the machine learning pipeline and image processing.


Today, farmers around the globe can more effectively protect their crops, allowing for a more sustainable crop protection strategy, greater yields, and economic prosperity.