CBL - Campus del Baix Llobregat

Projecte matriculat

Títol: Preliminary refinement and validation of an AI-based computer vision system for pest and disease diagnosis in citrus crops

Tutor/a o Cotutor/a: SALCEDO CIDONCHA, RAMON

Departament: DEAB

Títol: Preliminary refinement and validation of an AI-based computer vision system for pest and disease diagnosis in citrus crops

Data inici oferta: 23-09-2026 Data finalització oferta: 23-04-2027

Estudis d'assignació del projecte:
GR ENG CIEN AGRONOM

Lloc de realització:
EEABB

Segon tutor/a extern: ZHIHONG ZHANG

Paraules clau:
Artificial intelligence, Computer vision, Citrus crops, Plant diagnosis, Pests and diseases, Precision agriculture

Descripció del contingut i pla d'activitats:
It will focus on the preliminary refinement and validation of an artificial intelligence-based computer vision system for the diagnosis of pests and diseases in citrus crops. The work builds on previous activities carried out by the student at the Shanghai Institute of Technology (SIT), in collaboration with Prof. Zhihong Zhang and his research group. During this previous stage, the student participated in the preliminary testing of an AI system capable of analysing images of plant symptoms and providing a possible diagnosis. The proposed TFG will continue this work under new experimental conditions and with a specific focus on citrus crops. First, the methodology used to interact with the artificial intelligence system will be defined and standardized, including image acquisition procedures, prompt structure and criteria for evaluating the generated responses. A set of images representing different pests, diseases and visible symptoms affecting citrus plants will then be compiled. Part of the image dataset will be obtained from controlled or greenhouse conditions, while additional images may be obtained from reliable external sources when necessary. Each image will be associated with a reference diagnosis in order to establish the ground truth used for validation. The artificial intelligence system will subsequently analyse the selected images following the predefined methodology. Its responses will be compared with the reference diagnosis to identify correct classifications, misclassifications and recurrent diagnostic errors. Based on these results, the interaction methodology will be progressively refined, particularly the structure of the prompts and the contextual information provided to the system. A subsequent validation stage will be carried out using images acquired under outdoor or real crop conditions in Spain. This phase will allow a preliminary assessment of the robustness of the system under more variable environmental, illumination and image acquisition conditions. The performance of the system before and after refinement will be analysed and compared. The study will finally identify the main capabilities, limitations and sources of error of the proposed approach. The results will provide an initial assessment of the potential use of artificial intelligence and computer vision as support tools for image-based plant health diagnosis in citrus production.

Orientació a l'estudiant: Basic knowledge of crop production, plant health and agricultural engineering is recommended. Interest in artificial intelligence, computer vision and digital agriculture is desirable. The student should be available for experimental image acquisition and field or greenhouse activities when required. Basic skills in data organization and analysis will also be useful.