Leaf diseases represent a major threat to agricultural productivity and phytosanitary management, especially due to their rapid spread in crops. Manual plant inspection, the traditional method for detecting this type of affections, has several limitations: it is slow, subjective and difficult to scale up to large crop areas. Moreover, when symptoms are visible, in many cases it may be too late to apply effective control measures, increasing the risk of yield losses.
In this context, the FresQia demonstrator presents a proof-of-concept for automatic detection of foliar symptoms using computer vision and quantum computing, with the aim of supporting automated agronomic crop monitoring.
The proposed solution implements automatic plant condition monitoring using computer vision systems, integrating quantum computing techniques for the analysis of complex visual patterns. The approach compares classical models with Quantum Convolutional Neural Networks (QCNN), exploring the potential of hybrid models in precision agriculture applications.
Applied methodology
The developed system is structured in three main stages.
The first stage consists of capturing images in a controlled greenhouse growing environment, taking advantage of stable lighting conditions that allow obtaining high quality images. The images are taken over a line of potted crops with typical plant spacings, which facilitates individual plant identification and reduces leaf overlap. Even so, the system must face challenges inherent to these environments, such as overlap between leaves or the presence of background elements, such as weeds or pests.
The second stage corresponds to the automatic detection of leaves within each image. This uses a single-stage object detection architecture widely used in machine vision applications for its balance between accuracy and efficiency. The system automatically generates clippings of each detected leaf, scaled to 16×16 pixels from the original image. This process allows isolating the relevant information and optimizing the data that will later be processed by the classifier.
The third stage corresponds to the classification of the leaves, in order to determine whether the leaf is healthy or shows symptoms of leaf disease. This classification is performed using a Quantum Convolutional Neural Network (QCNN), whose results are compared with those obtained using a classical convolutional neural network (CNN).
Experiments were performed using a quantum emulation platform on classical HPC infrastructures, developed and maintained by CTIC. The QUTE platform is available to the research community and allows the execution and extension of the experiments presented, upon access request.
Results achieved
Preliminary results show a good performance of the system under favorable lighting conditions and when leaf symptoms are clearly visible.
However, some limitations are identified in more complex situations, such as in cases of leaf overlapping or when leaves are partially hidden in the image. The work analyzes these limitations and raises possible improvements to increase the robustness of the system in real agricultural environments.
Impact and application potential
The aim of this approach is to favor the early detection of foliar affections, which can contribute to reduce crop losses and decrease the dependence on pesticides, thus improving the sustainability of agricultural production.
In this study, the system has been applied to the analysis of strawberry leaves grown in greenhouses, although the methodology could be extended to other crops with visible leaf symptoms.
From a technological point of view, the work validates the feasibility of hybrid-quantum solutions for agricultural image analysis. The proposed architecture is compact and efficient, requiring less training data and lower memory consumption than many classical models, which facilitates its possible integration into agricultural devices and field monitoring systems.
In addition, the approach presents a high potential for scalability, with possibilities for continuous improvement as algorithms and quantum hardware evolve.