HybNet: A hybrid deep models for medicinal plant species identification
Real-time plant species detection plays an important role in fields ranging from medicine to biodiversity conservation. Images captured under unconstrained environments, scale variations, different lighting conditions, leaf orientation, complicated backdrops, and leaflet structure make plant species...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-06-01
|
| Series: | MethodsX |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2215016124005776 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Real-time plant species detection plays an important role in fields ranging from medicine to biodiversity conservation. Images captured under unconstrained environments, scale variations, different lighting conditions, leaf orientation, complicated backdrops, and leaflet structure make plant species recognition rigorous and time-consuming. Our study addresses this challenge by introducing three pioneering hybrid models, seamlessly integrating the strengths of convolution neural networks. In the first model, two deep learning models such as VGG16 and MobileNet are fused to extract features. Then, the extracted features are subjected to KNN classifier achieving an impressive 85.85 % accuracy, while the second model adopts MobileNet in conjunction with ResNet50 for feature extraction which is further classified using a deep learning classifier to achieve 88 % accuracy. The third model incorporates MobileNetV2 with the Squeeze and Excitation (SE) layers for the classification tasks. Our research highlights the immense potential of modern image processing techniques and deep learning models in comprehending and safeguarding the earth's diverse plant species. The experiments are carried out on self-created medicinal plant datasets captured in real-time conditions. From the experimentations, it is observed that hybrid model 3 reflects an improved performance of 94.24 % by utilizing recalibration efforts compared with the other two hybrid models. • One of the significant contributions of the study lies in a focused emphasis on feature enhancement achieved through the utilization of hybrid models majorly to enrich the features. • The feature scaling model incorporated in hybrid model 3 exhibits a superior and better performance demonstrating higher accuracy compared to the other models presented in this work. • The deebp learning models are trained and tested on the small dataset yet achieved good accuracy. |
|---|---|
| ISSN: | 2215-0161 |