Multi-Task learning with attention mechanism convolutional block attention (CBAM) for botanical image recognition / Joumana Aicha Ben Attalah

Public ISBD Titre : Multi-Task learning with attention mechanism convolutional block attention (CBAM) for botanical image recognition Type de document : document multimédia Auteurs : Joumana Aicha Ben Attalah, Auteur ; Farah Hadj, Auteur ; Nardjes Hamini, Directeur de thèse Editeur : Laghouat : Université Amar Telidji - Département d'informatique Année de publication : 2026 Importance : 48 p. Accompagnement : 1 disque optique numérique (CD-ROM) Note générale : Option : Artificial intelligence and data science Langues : Anglais (eng) Mots-clés : Plant recognition  Image classification  Computer vision  Fine-Grained classification  Explainable AI  Attention mechanisms  Lightweight deep learning Résumé : Plant species classification has become an essential task in the fields of agriculture, artificial intelligence,and computer vision due to its important role in plant monitoring, agricultural productivity enhancement,and biodiversity preservation.Nevertheless,manual plant identification remains a dificult and time-consuming process because many plant species share highly similar visual characteristics,suchasshape,color,and texture, making accurate recognition challenging. This project proposes an intelligent deep learning model based on Multi-Task Learning (MTL) combined with the Convolutional Block Attention Module(CBAM)toenhance plant analysis and classification from images.The proposed approach simultaneously performs four related tasks:plant localization with in the image,plant species classification in to 30 categories,color classification in to 16 categories,and shape classification in to 5 categories. By learning these tasks jointly,the model is able to extract shared visual representations that improve the overall performance and robustness of the system. The proposed model relies on MobileNetV2 as the backbone network for feature extraction, while transfer learning and progressive training strategies were adopted to improve adaptation to the plant domain.Furthermore,Grad-CAM was employed to visualize and interpret the model's predictions by highlighting the most influential image regions involved in the classification process,thereby improving the explainability of the model. Experimental results demonstrated the effectiveness of the proposed approach,achieving a classification accuracy of 97%,a precision score of 96.9%, a weighted F1-score of 94.06%, and a low loss value of 0.2.These findings confirm the model's ability to accurately and reliably classify plant species and identify their visual characteristics,highlighting its potential for future applications in smart agriculture and automated plant monitoring systems. note de thèses : Mémoire de master en informatique Multi-Task learning with attention mechanism convolutional block attention (CBAM) for botanical image recognition [document multimédia] / Joumana Aicha Ben Attalah, Auteur ; Farah Hadj, Auteur ; Nardjes Hamini, Directeur de thèse . - Laghouat : Université Amar Telidji - Département d'informatique, 2026 . - 48 p. + 1 disque optique numérique (CD-ROM). Option : Artificial intelligence and data science Langues : Anglais (eng) Mots-clés : Plant recognition  Image classification  Computer vision  Fine-Grained classification  Explainable AI  Attention mechanisms  Lightweight deep learning Résumé : Plant species classification has become an essential task in the fields of agriculture, artificial intelligence,and computer vision due to its important role in plant monitoring, agricultural productivity enhancement,and biodiversity preservation.Nevertheless,manual plant identification remains a dificult and time-consuming process because many plant species share highly similar visual characteristics,suchasshape,color,and texture, making accurate recognition challenging. This project proposes an intelligent deep learning model based on Multi-Task Learning (MTL) combined with the Convolutional Block Attention Module(CBAM)toenhance plant analysis and classification from images.The proposed approach simultaneously performs four related tasks:plant localization with in the image,plant species classification in to 30 categories,color classification in to 16 categories,and shape classification in to 5 categories. By learning these tasks jointly,the model is able to extract shared visual representations that improve the overall performance and robustness of the system. The proposed model relies on MobileNetV2 as the backbone network for feature extraction, while transfer learning and progressive training strategies were adopted to improve adaptation to the plant domain.Furthermore,Grad-CAM was employed to visualize and interpret the model's predictions by highlighting the most influential image regions involved in the classification process,thereby improving the explainability of the model. Experimental results demonstrated the effectiveness of the proposed approach,achieving a classification accuracy of 97%,a precision score of 96.9%, a weighted F1-score of 94.06%, and a low loss value of 0.2.These findings confirm the model's ability to accurately and reliably classify plant species and identify their visual characteristics,highlighting its potential for future applications in smart agriculture and automated plant monitoring systems. note de thèses : Mémoire de master en informatique

---

1 (1 - 1 / 1) Par page : 25 50 100 200