Hamed Mehdipoor
Spectro-AG BV
Contact by mail1 History: Spectro-AG BV is a knowledge-based company that focuses on research and development of solutions based on the application of multi/hyperspectral data and artificial intelligence (AI) for precision agriculture. We are developing software and hardware systems to collect and process big spatio-spectral data from farms in (semi) real-time. Our software systems fuse, analyze and process image data from different sensors and platform to generate information that fits farm management models of the user. We believe the automatic generation of such information support decision-makers in agriculture. In particular, Spectro-AG BV company possesses necessary skills and experience to successfully deliver a high-quality product – this claim can be substantiated by a number of past and ongoing projects in the field of remote sensing and AI (examples include automatic weed detection in agricultural fields, mildew detection in wine grapes, etc.). 2 Main domains Crops use the limited ecosystems resources while the climate is dramatically changing in the Netherlands. This use can diminish the ecological and economical quality of the countries in the country in the long term. Grass farmers often take more prescriptions in the form of fertilization to keep the quality (e.g., crude protein content) in their crops in a high condition and to deliver a high production quota. As a result, preserving crops quality in the ecosystem is a sliding scale. Too little fertilization will drain the natural resources of its nutrients, and too much use of fertilization will disrupt the natural ecosystem. Unbalanced use of fertilization has consequences for the long-term economic vitality of crops. Therefore, regular and timely information about crops quality is highly demanded optimizing grass farming practices. The current approach to sample the grass and send it to the laboratory for analyzing is labour-intensive and time-consuming and expensive if many samples are required. In this respect, HyperSlit drone is a non-destructive and timely solution to produce information about grass quality. Our aim is to scale up the solution to helps farmers to produce high-quality grass at a national scale. Further, organizations that provide fertilization and that assess the effect of fertilizers can use the information to take timely action. We have developed a compact drone hyperspectral (hardware and software) solution, called HyperSlit, to acquire, qualify and process big data about crops in (semi) real-time. Unlike available drone hyperspectral system in the market that are expensive (~100K Euro) and complicated, our HyperSlit (~15K Euro) is affordable and automatic for generating georeferenced maps about crops traits such as protein and fibre content in grassland. This information is immediately available after flights over farms and it can supports farmers and contractors to improve their management practices such as grass manuring and mowing as well as soil treatments. 3 Projects of the company 3.1 Automatic detection and monitoring downy mildew in vineyards in the Netherlands using satellite and drone images: in this project, we used machine learning and data mining methods to analyze and model satellite and drone images to localize the grape trees affected with downy mildew (Figure 3). As the climate is getting warmer, wine production is getting more and more popular in the Netherlands. However, farmers are facing a specific infection in trees because of the raining condition in the Netherlands. The information that we produced helps the farmers to not sprayer herbicides everywhere on the farm, but only over the infected tree. Figure 3. Classification of the vineyard for downy mildew 3.2 Automatic detection and spraying of weed in real-time using artificial intelligence: In this project, Spectro-AG BV processed and trained an object detection model that could detect a specific weed in multispectral images from the camera. In particular, we automatically retrieved multispectral images from iNaturalist databases to train the model. The model detects Bitter Dock weed in RGB camera images in real-time (Figure 2). We developed a software application that processes the live-stream video of a connected camera. When the model detects the bitter dock in the camera live-stream, the software triggers the sprayer module in the night time! Figure 2. Detected bitted dock weed in real-time 3.3 Monitoring and mapping grassland traits using satellite and drone images and artificial intelligence: As the Research and development leader partner in this project we have collected and process multispectral and hyperspectral images collected from satellite and drones to estimate grassland traits including the protein level in the grass (Figure 1). In this project, we used the big data collected from the farms and trained an AI-based model that use satellite and UAV images as input and estimate the crude protein as output. Figure 1. Protein map of the grassland 4 Our Key persons for this project: Our international team consists of young but experienced researchers and developers that have background and competence in precision agriculture, remote sensing, robotics, big data, artificial intelligence. These motivate us to research and develop knowledge-based solutions for real agricultural and environmental problems in the Netherlands. In particular, the below team members and skills could uniquely be positioned to succeed in proofing the application of our HyperSlit solution for grassland monitoring: • Hamed Mehdipoor (PhD): crops phenology, artificial intelligence and Hyperspectral sensing, big data • Manuel Garcia (PhD): precision agriculture, cloud computing, smart sensors and application, • Tatjana Kuznecova (MSc): Environmental assessment, image processing, biochemistry, soil geothermal • Andrej Pistak (BSc): Software development, machine learning, artificial intelligence, wireless communication