FPGA Implementation of Cotton Leaf Disease Detection Using CNN

The work done here focuses on developing an innovative method for identifying plant diseases using Convolution Neural Networks (CNN) on the PYNQ FPGA platform. One of the advantages of employing an FPGA is that it accelerates the inference of CNNs, thereby improving the effectiveness of plant disease detection. Considerations like image dimensions and learning rates are taken into account while designing the CNN architecture to be compatible with the PYNQ FPGA. The design is experimentally evaluated in Python yielding a training accuracy of 96% in simulated data, while accuracy for the hardware implementation remains to be established. A training algorithm which yields optimal values for hyper parameters is devised so that accuracy, F1 scores, and other performance measures are enhanced. CNN led visualizations also include the bounding boxes of the leaves showing features like colours, shapes, and textures that are used to identify moderate to heavily infested plants. Apart from these, there is also consideration of resource constrictions on the FPGA to ensure that the design is flexible for different hardware applications. Moreover, the system demonstrates accuracy by drawing bounding boxes on the CNN output images of the leaves which depicts the health status of the plant. This includes eyes plant disease classification which utilizing FPGA technology gives a great improvement in performance.