Protein is made up of a variety of molecules that are required by living organisms, such as enzymes, hormones, and antibodies. In step 2, the max-pooling layer and the convolutional layer evaluate the input data to create the finest feature map F1, which is half the image size in both horizontal and vertical directions. The full feature is then retrieved in step 2 using the max pooling layer and the residual block at the proper resolution. In this paper, we introduce Di-Fuzzy CNN (Fuzzy Convolutional Neural Network with Dingo optimizer), a novel technique for predicting protein activities that incorporates two types of information they are protein sequence and protein structure. We extract diverse features at different scales utilizing convolutional neural networks to provide comprehensive information for feature segmentation. To handle a variety of uncertainties in feature selection and produce segmentation results that are more dependable, fuzzy logic modules are employed. Finally, we employ Dingo optimization to boost the suggested method's effectiveness and speed in order to produce the best outcomes. Using a variety of datasets, the suggested model has been tested (HSSP, PDB, UGR14b, DSSP). Tests demonstrate that our approach can decrease FPR, increase protein structure accuracy, decrease prediction time, and increase TPR for feature selection. Our predictive model performs better than most state-of-the-art techniques.