Improvedinception-V3 model forapple leaf disease classification

Abstract

Apple,  a  nutrient-rich  fruit belonging  to  the genus Malus, is  recognized  for its fiber, vitamins, and antioxidants, giving health benefits such as improved digestion and reduced cardiovascular disease risk. In Indonesia, the soil and climate  create  favorable  conditions  for  apple  cultivation.  However,  it  is essential  to  prioritize  the  health  of  the  plant.  Biotic  factors,  such  as  fungal infections   like   apple   scabs   and   pests,   alongside   abiotic   factors   like temperature  and  soil  moisture,  impact  the  health of  apple  plants.  Computer vision, specifically convolution neural network (CNN)inception-V3, proves effective in aiding farmers in identifying these diseases. The output layer in inception-V3   is   essential,   generating   predictions   based   on   input   data. For  this  reason,  in  this  paper,  we  add  an  output  layer  in inception-V3 architecture  to  increase  the  accuracy  of  apple  leaf  disease  classification. The  added  output  layers  are  dense,  dropout,  and  batch  normalization. Adding  a  dense  layer  after  flattening  typically  consolidates  the  extracted features  into  a  more  compact  representation.  Dropout  can  help  prevent overfitting  by  randomly  deactivating  some units  during  training.  Batch normalization   helps   normalize   activations   across   batches,   speeding   up training  and  providing  stability  to  the  model.  Test  results  show  that  the proposed   method   produced   an   accuracy   of   99.27%   and   can   increase accuracy   by   1.85%   compared   to inception-V3.   These   enhancements showcase  the  potential  of  leveraging  computer  vision  for  precise  disease diagnosis in apple crops.