Efficient Siamese Pyramid Attention Network with Fusion Autoencoder and Hypercube Gan for Detection of Pesticide Residue

Excessive pesticide use in agriculture boosts productivity but raises health risks, complicating pesticide residue detection in fruits and vegetables through hyperspectral imaging. Hence, a novel “Siamese Pyramid Attention with Fusion Autoencoder and Hypercube GAN” is introduced. In the existing segmentation process, it poses challenges due to spectral similarities between pesticide residues and healthy fruits and vegetables. Thus, a novel “Siamese Dual Path Network with Feature Pyramid Attention (DPN-FPA)” is employed, to enhance segmentation by reducing errors by distinguishing residues effectively. Furthermore, temporal dynamics in pesticide residue analysis led to chemical composition variations, necessitating improved unmixing techniques for accurate feature extraction. So, a novel “Sparse Capsule Spatial-Spectral Fusion Autoencoder (SC-SSFA)” is introduced, which seamlessly incorporates spatial and spectral data, enhancing accuracy in pesticide residue detection. Finally, detection faces trade-offs between higher spatial and higher spectral resolutions. Therefore, a novel “Hypercube Conditional Generative Adavector Network (HCGAN)” is implemented, which optimally balances spatial and spectral resolutions using a conditional generative adversarial network architecture thereby enhancing pesticide residue detection in fruits and vegetables. The proposed method demonstrates excellence in segmentation, feature extraction, and detection by achieving a high accuracy of 0.99, precision of 0.98, recall of 0.98, and low mean square error (MSE) of 2.4%.