Journal DOI 10.68956/journal.human-performance-futures

Creative Commons source reading · Data quality and algorithm evaluation

Hybrid fusion of E-nose and computer vision using optimized deep learning and machine learning for robust plant leaf recognition

Bohlol P; Dizavandi MHS; Mohtasebi SS; Omid M

Originally published in Sci Rep · 2026-06-15

From the article

Selected source text. JournalOS removes markup and shortens excerpts where indicated. The full article provides the complete context.

Abstract

AbstractThe fusion multi-sensory system with optimized deep learning and machine learning algorithms appeared to synergize difficult paradigms in precision agriculture and boost recognition of various plant species. In this study, an electronic nose (E-nose) system with eight MOS sensors and a computer vision platform were designed for constructing a large-scale data set of 26 various species of plant leaves collected during 2 years under different environmental conditions. After preprocessing functions, the E-nose data set consisted of 15,600 data points, while two-phase data augmentation expanded the image data set to 156,000 samples. Multiple machine learning models, including MLPs and 7 machine learning algorithms, were executed and optimized based on the E-nose dataset, where KNN with 97% accuracy and 5 ms response time per…

Results

Leaves of plants have high similarity in shape, color, texture, and odor that stretch back to the condition of foster, weather of environment, and breeds. 26 leaves were selected for integrating the E-nose sensor with the machine vision system. Data was analyzed with machine learning and deep learning algorithms. The combination of deep learning and machine learning algorithms provokes the detection of the appearance and aroma of leaves and makes it possible to also compare the performance of these platforms. All models were trained and tested on Google Colab using an A100 GPU. Reported training and inference times were measured in this environment. While this setup allows for reproducible benchmarking, the models have not yet been deployed or tested on real edge devices. Future work will investigate deployment on web-based or mobile…

Discussion

Optimizing the model to increase performance is one of the important stages of developing a framework. Follow that, optimizing VGG16 and AlexNet by appending specific layers showed a positive impact58, while improving architecture with integrating specific layers to ResNet18 yielded notable results59. Similarly, robust algorithms have been developed by feeding different blocks of data sets that are specific60. Beyond optimization architectures, developing multi-sensor systems has also been described behind optimizing algorithms to exploit full feature extraction61. Most previous studies have encountered challenges in selecting the method of integration or have focused on optimizing specific hyperparameters. In addition, in most investigations, feature extraction was typically performed by either processing the two methods separately…

Editorial reading guide

  • What data created and tested the model, and was evaluation independent?
  • What is the cost of a false positive, false negative or out-of-distribution decision?