Research

Authors: S. S. Swapnil, S. K. Sarker, A. B. Dibya, M. T. Islam, M. A. T. Roni, K. Muhammad

Abstract — Accurate classification of crop damage is essential for improving agricultural productivity and sustainability. It enables timely interventions, loss reduction, and optimized resource utilization. It also supports targeted pest control and disease management strategies. Despite its importance, datasets and models are scarce for binary classification of damaged versus non-damaged crops. To address this, we conducted an extensive study on crop damage classification using deep learning, focusing on the challenges posed by imbalanced datasets common in agriculture. We began by preprocessing the Consultative Group for International Agricultural Research (CGIAR) dataset to enhance data quality and balance class distributions. We created the new Crop Damage Classification (CDC) dataset tailored for binary classification of damaged versus non-damaged crops, serving as an effective training medium for deep learning models. Using the CDC dataset, we benchmarked various state-of-the-art models to evaluate their effectiveness in detecting crop damage. We introduced a custom model named Light Crop Damage Classifier (LightCDC) for computationally efficient crop damage classification to optimize performance in resource-constrained precision agriculture. Leveraging the depth channel shuffling technique of ShuffleNetV2, we reduced parameters from 1.40 million to 1.13 million while achieving an accuracy of 89.44\%. LightCDC outperformed existing classification and ensemble models regarding model size, parameter count, inference time, and accuracy. Furthermore, we tested LightCDC under adverse conditions like blur, low light, and fog, validating its robustness for real-world scenarios. Thus, our contributions include a refined dataset and an efficient model tailored for crop damage classification, essential for timely interventions and improved crop management in precision agriculture.

Abstract — will update soon

Abstract — The volatility of cryptocurrency prices presents significant challenges for accurate prediction. This paper introduces a novel machine learning framework using a custom Physics-Informed Neural Network (PINN) to predict the closing prices of cryptocurrencies like Bitcoin (BTC) and Ethereum (ETH). The proposed PINN model is benchmarked against optimized versions of Random Forest and k-Nearest Neighbor (KNN) models. The dataset, spanning five years, undergoes comprehensive pre-processing, including stationarity testing using the Augmented Dickey-Fuller (ADF) method, detrending through differencing transformation, and Min-Max normalization. Performance metrics such as Mean Squared Error (MSE), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE) are employed to assess model accuracy. Results indicate that the custom PINN outperforms traditional models, achieving RMSE values of 0.20201 for BTC and 0.17561 for ETH, representing improvements of 22.70% and 21.59% over KNN and Random Forest for BTC, and 27.27% and 25.37% for ETH, respectively. The study underscores the potential of incorporating physics-informed approaches into machine learning models to enhance the predictability of highly volatile financial markets.