In the heart of Morocco, a groundbreaking development is set to revolutionize the way we approach citrus farming. Manal El Akrouchi, a researcher at the College of Agriculture and Environmental Sciences, University Mohammed VI Polytechnic (UM6P), has led a team to create an AI-based framework that promises to transform early detection and segmentation of green citrus fruits in orchards. This innovation, published in ‘Smart Agricultural Technology’ (Intelligent Agricultural Technology), could significantly impact yield prediction, resource management, and decision-making in the agricultural sector.
The framework combines Multiscale Vision Transformers version 2 (MViTv2) with Cascade Mask R-CNN, a powerful duo that addresses the challenges of detecting and segmenting tiny green citrus fruits in dense orchards. Unlike traditional methods that rely on close-up images, this new approach extends its focus to full-tree images, providing a more comprehensive view of the orchard. “By using full-tree images, we can capture the complexity of real-world orchard settings, where dense foliage and small fruits make detection challenging,” El Akrouchi explains. This dual-image strategy—using close-up images for training and full-tree images for testing—enhances the model’s ability to perform accurately in practical scenarios.
One of the standout features of this framework is its innovative image-slicing method. High-resolution full-tree images are broken into smaller parts to capture finer details, significantly boosting detection accuracy. This technique was tested on a diverse dataset featuring three varieties of citrus orchards: Nules grafted on Volka, Sidi Aissa grafted on Volka, and Orogrande grafted on sour orange. The results were impressive, with the MViTv2_L backbone achieving a mean Average Precision (mAP) of 72.97% for bounding boxes and 84.40% for masks. The image-slicing technique further enhanced fruit detection, achieving an R2 value of up to 0.81 for fruit counting.
The implications of this research are vast. For citrus farmers, this technology means more accurate yield predictions, better resource management, and timely decision-making. “This dual-image method, paired with advanced segmentation and detection technologies, marks a significant step forward for agricultural robotics and precision farming,” El Akrouchi notes. The ability to detect and segment fruits at an early stage can lead to more efficient harvesting, reduced labor costs, and improved overall productivity.
As the world continues to embrace precision farming and agricultural robotics, this research paves the way for future developments. The integration of AI and advanced imaging techniques into agricultural practices could lead to smarter, more sustainable farming methods. This breakthrough not only benefits citrus farmers but also sets a precedent for other crops and agricultural sectors, potentially transforming the way we approach farming on a global scale.