Publication

Abstract : This paper presents a Deep Residual Squeeze Excited Region Convolutional Network for the precise classification of plant diseases. The proposed model combines deep residual learning, adaptive feature recalibration, and region-focused convolution to improve accuracy and efficiency in disease detection. Leveraging datasets like PlantVillage dataset, Citrus Leaf Disease image dataset, and Coffee Leaf diseases dataset undergo extensive preprocessing, including normalization, data augmentation, resizing, noise cancelation, and contrast enhancement to ensure high-quality inputs. Feature extraction is enhanced by multihead self-attention mechanisms with transformers, generating a detailed, high-dimensional image representations. To manage the complexity of the processed data, statistical techniques such as mean, variance, skewness, kurtosis, and standard deviation are employed, capturing intricate details that aid in classification. The proposed model utilizes squeeze-and-excitation module to emphasize essential features within the images, refining the focus on disease-relevant characteristics. Experimental evaluations use multiple metrics, showcasing the model’s superior accuracy of 98.53%, precision of 97.23%, recall of 96.60%, F1 score of 96.91% and specificity of 96.63%. This rigorous evaluation confirms the model’s practical viability for high-stakes agricultural applications.