Simple Vision System for Apple Varieties Classification

AbstractEvery variety of apple has its particular physical characteristics, which are affected by different pre-harvest factors. Manual classification of these varieties by human labor has several weaknesses, such as the inconsistency, subjectivity, fatigue and different accuracy due to different level of experience of the inspector. This study was aimed to design and evaluate a simple computer-based vision system for recognizing and grading several varieties of apples based on their physical characteristics. Images of apples were taken and were used as training data with different algorithms to extract the particular characteristics of each variety, such as color and shape. The extracted Hue color channels and contour vector were recorded as the reference data and were used to recognize the similar characteristic of those images from the testing data group. The k-nearest neighbors algorithm was used to decide whether an apple belongs to a particular variety. The results show that the recognition rate based on color only was between 84–97% and it was between 5–77% it is based on the shape only. Rotating the image significantly increases the recognition rate (to be between 5 - 69% based on the shape only). Moreover, combining both color and shape characteristics significantly improves the recognition rate.Keywords: apple’s varieties classification, color signatures, combined color-morphology signatures, morphology signature, vision system AbstrakSetiap jenis buah apel memiliki penciri fisik spesifik, yang dipengaruhi oleh berbagai faktor pra-panen. Teknik klasifikasi manual memiliki banyak kelemahan, antara lain adalah subjektifitas, ketidakkonsistenan, kelelahan fisik dan psikologis, serta tingkat pengalaman dari petugas yang melakukannya. Tujuan studi ini adalah melakukan proses desain dan pengujian suatu sistem visi sederhana berbasis komputer untuk mengenali dan mengklasifikasi berbagai jenis buah apel berdasarkan penciri spesifiknya. Citra buah apel dari sampel latih diproses dengan berbagai algoritma untuk mengekstraksi berbagai parameter pencirinya, yaitu parameter warna dan bentuk. Informasi histogram kanal warna Hue dan vektor kontur hasil ekstraksi kemudian disimpan sebagai data referensi dan digunakan sebagai pembanding terhadap parameter serupa dari citra data uji. Keputusan diambil menggunakan algoritma K-Nearest Neighbors. Hasil menunjukkan bahwa laju pengenalan berbasis fitur tunggal warna berkisar antara 84–97%, sementara berbasis fitur tunggal morfologi berkisar antara 5–77%. Perubahan orientasi sampel sebagai data training akan meningkatkan laju pengenalan berbasis fitur tunggal morfologi secara signifikan, yaitu dari 5% menjadi 69%. Penggabungan dua fitur penciri warna dan morfologi dapat meningkatkan laju pengenalan lebih baik lagi.Kata Kunci: klasifikasi jenis buah apel, penciri warna, penciri morfologi, gabungan penciri warna dan morfologi, sistem vis