Development and Validation of a Multispectral Vision System for In-Situ Detection of Pesticide Residues on Agricultural Produce

Abstract

In response to the pressing demand for rapid and accurate detection of pesticide residues on the surface of fruits and vegetables, this study proposes an innovative visual detection system based on multispectral fusion. The system integrates an improved deep convolutional neural network (DCNN) with a Transformer architecture, enabling precise feature extraction from multispectral images. By leveraging the complementary strengths of these models, the system achieves enhanced classification accuracy and robustness against environmental variations. Furthermore, an adaptive weighted fusion algorithm, combined with improved kernel principal component analysis (KPCA), is employed to optimize the quality of spectral data fusion. This approach effectively mitigates the impact of noise and enhances the discriminative power of the features. To address the challenge of real-time detection, a parallel computing architecture based on CUDA and a heterogeneous hardware configuration scheme are designed. This implementation significantly accelerates the processing speed, enabling rapid analysis of large-scale datasets. Additionally, system stability is enhanced through redundant hardware design, intelligent fault-tolerant mechanisms, and optimized data transmission protocols. These measures ensure reliable operation under diverse environmental conditions, making the system suitable for field applications.