Open world plant image identification based on convolutional neural network

Sun

Computational Intelligence and Neuroscience

2017

511

212

Automatic and accurate estimation of disease severity is essential for food security, disease management, and yield loss prediction. Deep learning, the latest breakthrough in computer vision, is promising for fine-grained disease severity classification, as the method avoids the labor-intensive feature engineering and threshold-based segmentation. Using the apple black rot images in the PlantVillage dataset, which are further annotated by botanists with four severity stages as ground truth, a series of deep convolutional neural networks are trained to diagnose the severity of the disease. The performances of shallow networks trained from scratch and deep models fine-tuned by transfer learning are evaluated systemically in this paper. The best model is the deep VGG16 model trained with transfer learning, which yields an overall accuracy of 90.4% on the hold-out test set. The proposed deep learning model may have great potential in disease control for modern agriculture.

Section: Resultssupporting

confidence: 89%

Section: Related Workmentioning

confidence: 99%

Automatic Image-Based Plant Disease Severity Estimation Using Deep Learning

Sun

Computational Intelligence and Neuroscience

2017

511

212

“…Activation functions such as ReLU and Leaky ReLU introduce non-linearity, enabling CNNs to solve complex problems [24]. Leaky ReLU, defined as f(x) = max(0.01x, x), modifies ReLU by replacing the horizontal line for x < 0 with a non-zero, non-horizontal line [25]. However, the exploding gradient problem is not solely dependent on the learning rate or specific to Leaky ReLU.…”

Section: Convolutional Neural Networkmentioning

confidence: 99%

A Recurrent Feature Iterative Network For Suspect Identification using Convolutional and Long Short Term Memory Approach ⋆

Shree,

Mohapatra,

Suwanwiwat

et al. 2024

Preprint

In suspect identification systems, facial features play a crucial role in recognising individuals. However, the challenge lies in sustaining the accuracy of the system over a long period of time, ensuring that it remains consistently high, reliable, and effective. This research introduces a novel lightweight model that requires low trainable parameters, a significantly smaller number than pre-trained models, which use millions of trainable parameters. The newly proposed recurrent feature iterative network integrates a convolutional neural network and long short-term memory in a single structure to synthesise diverse images and to effectively extract facial features. The long short-term memory-based feature-recurrent system demonstrates a significant improvement in accuracy when tested on the augmented reality face database, enhanced extended Yale B, Cohn-Kanade and extended Yale B datasets, achieving encouraging accuracy rates of 99.23%, 99.16%, 98.99%, and 97.68%, respectively. These accuracies, in general, outperform traditional baselines of 68.65%. This research advances the field by providing an innovative solution to enhance suspect identification systems through advanced facial image feature extraction, resulting in significantly improved accuracy rates.

Eng. Technol. Appl. Sci. Res.

“…Authors in [6] conducted research regarding plant image classification tasks, for instance classifying maize plants and weeds by utilizing segmented images at initial stages of growth, with a training accuracy of 97.00%. Image processing and identification have been the subjects of [7][8]. Weed identification is challenging to ambiguous crop constraints and differing rocky or sandy identities, and long established classification techniques are most likely to fail in this task [9].…”

Section: Introductionmentioning

confidence: 99%

Deep Convolutional Neural Network Architecture for Plant Seedling Classification

Kundur

Mallikarjuna

2022

Weed control is essential in agriculture since weeds reduce yields, increase production cost, impede harvesting, and degrade product quality. As a result, it is indeed critical to recognize weeds early in their vegetation cycle to evade negative impacts to crop growth. Earlier traditional methods used machine learning to determine crops along with weed species, but they had issues with weed detection efficiency at early growth stages. The current work proposes the implementation of a deep learning method that provides accurate results for precise weed recognition. Two different deep convolution neural networks have been used for our classification framework, namely Efficient Net B2 and Efficient Net B4. The plant seedlings dataset is utilized to investigate the proposed work. The evaluation metrics average accuracy, precision, recall, and F1-score were used. The findings demonstrate that the proposed approach is capable of differentiating between 12 species of a plant seedling dataset which contains 3 crops and 9 weeds. The average classification accuracy and F1 score are 99.00% for our Efficient Net B4 model and 97.00% for the Efficient Net B2. In addition, the proposed Efficient Net-B4 model performance is compared to the one of existing models on the plant seedlings dataset and the results showed that the proposed model Efficient Net B4 has superior performance. We intend to detect diseases in the identified plant species in our future research.