Plant diseases have important consequences for livelihoods and economies, both on local and global scales, whereby the spread of plant pathogens can lead to high levels of damage to agricultural productivity. In this regard, deep learning (DL) has evolved as a promising remedy. Nevertheless, the level of diversity of plant species still presents a constant challenge to effective plant disease classification. This research proposed a two-stream convolution operator combining bidirectional long short-term memory (BiLSTM) for effective feature mapping and learning. Moreover, a transformer network (TransNet) is constructed based on sequential learning techniques (SLT) of long short-term memory (LSTM), bidirectional LSTM (BiLSTM), sequence-to-sequence, and gated recurrent unit to highlight important plant disease features. The network is trained on the PlantVillage dataset, which includes 38 species of plant diseases. The experimental results indicate that the proposed model achieves scores of 97.88%, 97.93%, and 97.62% for Accuracy, Precision, and Recall, respectively, along with a low training loss of 0.0696 and a minimal training time of 83.17 minutes. The proposed approach represents a significant improvement over these previous models and demonstrates the potential for the efficiency and accuracy of plant disease classification to be further improved, contributing to the realization of modern technology in the agricultural industry.
Artificial intelligence (AI) integration has become crucial in augmenting safety systems at train stations and along railway lines. It plays a crucial role in identifying and monitoring individuals, detecting hazardous items, classifying unusual movements, and recognizing obstructions on railway lines. Artificial intelligence’s capacity to swiftly and precisely analyze data enables systems to react to possible dangers promptly. The present study introduces a sophisticated railway track identification system that utilizes semantic segmentation methods and deep learning models to effectively distinguish between items and persons in close proximity to railway zones. This work combines ResNet-50 with five segmentation models: SegNet, U-Net, FCN-8, FCN-16, and FCN-32. Refinements to the model’s underlying architecture are intended to enhance its performance, while a dataset of 5,000 images from the internet is utilized for testing. This paper compares traditional and improved model architectures, including precise learning rate adjustment to maximize accuracy. Empirical findings demonstrate that incorporating ResNet-50 substantially enhances accuracy and precision. The FCN-16 + ResNet-50 model had a mean accuracy coefficient of 98.72%. Despite the increased computational parameters, the improved models paradoxically enable quicker processing across all models.
Smart waste classification and image retrieval are important for recycling, resource recovery, and circular-economy applications. This study presents a comparative deep learning framework for waste classification and content-based image retrieval (CBIR) under geometric transformations. Rather than proposing a single fused architecture, the study systematically investigates three key design factors: backbone architecture, classifier-head design, and loss-function selection. A custom waste image dataset comprising 1000 images across 10 waste categories was used throughout the experiments. The evaluated backbone architectures included ResNet18, ResNet34, and ResNet50, while the classifier heads consisted of a fully connected classifier, a DenseNet201-based classifier, and a VGG16-based classifier. In addition, three loss functions—cross-entropy, label smoothing, and ArcFace—were evaluated under identical experimental conditions. The results show that ResNet50 combined with a DenseNet201-based classifier and ArcFace achieved the highest classification performance, reaching a 98.4% accuracy and 98.3% F1-score. ArcFace also improved feature discriminability and enhanced retrieval effectiveness, with the best configuration achieving a 98.1% mean average precision (mAP). Robustness evaluation under rotation and scaling transformations demonstrated that the proposed framework maintained stable performance under geometric variations. Although ResNet50 achieved the highest accuracy, ResNet34 combined with a fully connected classifier and ArcFace provided a favorable balance between recognition performance and model complexity. These findings demonstrate that backbone depth, classifier-head design, and loss-function selection all play important roles in determining waste classification and retrieval performance.