ANALISIS KINERJA JARINGAN SYARAF BERBASIS SKIP CONNECTION UNTUK KLASIFIKASI HAMA SERANGGA

PERFORMANCE ANALYSIS OF SKIP CONNECTION-BASED NEURAL NETWORKS FOR INSECT PEST CLASSIFICATION

Abstract

Insect pests are a significant problem in food crop production. Identifying insects that threaten food crop production is an effort to overcome these pest problems. The proper identification process will be able to provide the right treatment solution according to the type of insect that is the problem. Artificial intelligence is a technology-based solution for more accurate identification, where human fatigue will likely cause identification errors. Artificial neural networks are algorithms in artificial intelligence capable of carrying out image-based classification tasks. Skip Connection is a layer in an artificial neural network that can improve the performance of a convolutional-type artificial neural network (CNN). DenseNet121 and ResNet50 are two CNN architectures that are pretty popular. Both have a Skip Connection layer with different variations. This research explores and analyzes the performance of two different Skip Connection architectures, DenseNet 121 and ResNet50, in handling insect pest image classification. The results were obtained using two different insect pest image datasets. DenseNet121 has a better performance than ResNet50. Using the balanced accuracy score metric, the performance of DenseNet121 versus ResNet50 is 0.6361:0.5053 and 0.8598:0.7017. The difference in the performance of Skip Connection on DenseNet121 compared to ResNet50 is ± 10% better in the image classification of two insect pest datasets.