Karadeniz, Alper TalhaCelik, YukselBasaran, Erdal2024-09-292024-09-2920231438-23771438-2385https://doi.org/10.1007/s00217-022-04168-8https://hdl.handle.net/20.500.14619/3842Walnuts are widely used, although they come in a variety of types and qualities. It is essential to choose the correct walnut variety with the necessary ecological characteristics to continue the production of walnut fruit, which has positive benefits on human health. Because planting a walnut garden is expensive and the harvesting process takes a while. However, since the colour and feel of walnut leaves are so similar, it can be challenging to tell them apart. Experts must devote a significant amount of time to differentiating walnut kinds, and morphological tests should be conducted. There are different studies in the literature for walnut variety differentiation. Nevertheless, those are studies conducted with the classification of a small number of walnut varieties or laboratory experiments. With the advancement of technology, deep learning techniques based on computers are now routinely utilized for leaf recognition. These technologies enable significant reductions in error rates, time saves, and cost. With a total of 1751 leaf pictures collected from 18 species of walnuts, a special walnut dataset was constructed for this study in order to identify walnut types from walnut leaves. To automatically classify the provided dataset, images are trained with residual block-based convolutional neural network architectures. Following the discovery of each image's deep features, the Atom Search Optimization algorithm was used to choose the most distinctive characteristics. Support vector machines (SVM) were used to classify walnut species with the new feature set created. The experimental studies of the proposed model based on Residual block and Atom Search optimization successfully categorised the walnut dataset with an accuracy rating of 87.42%.eninfo:eu-repo/semantics/closedAccessWalnut datasetResidual blockAtom search optimizationOptimization based feature selectionSupport vector machinesArticle10.1007/s00217-022-04168-82-s2.0-851416638457383Q2727249WOS:000881617300001Q2