Pixel-level rapid detection of aflatoxin B1 based on 1D-modified temporal convolutional network and hyperspectral imaging

Aflatoxin is a virulent toxic and cancerogenic substance, and it is widespread on the peanuts surface. This study uses a one-dimensional modified temporal convolutional network (1D-modified TCN) to detect aflatoxins at the pixel level with hyperspectral images. Then, the robustness of this method was verified from compressed image size and reduced spectral acquisition points, and feasible strategies were provided for accelerating model optimization training. The experimental results indicate that 1D-modified TCN is best under the image size of 63×47, the model training accuracy was 99.60 %, and the test accuracy was 99.26 %. Finally, the detection results of the four one-dimensional network models were visualized on three new peanut kernels. This method has improved the accuracy of aflatoxin detection and it will quickly promote the design and development of intelligent devices in detecting aflatoxin.