No abstract

Identifying juvenile grasses is especially difficult because their characteristics are greatly affected by growing conditions. One problem, which affects both computer‐based identification systems and written identification keys, is the inability of users to select the correct alternatives for some morphological characters or structures. This study was designed to determine, for each selected character, whether individuals without training in plant identification could identify each character state correctly as often as trained individuals, and to determine if combining poorly distinguished character states would increase the value of a character for identification of the selected species. seeds or young seedlings of 18 grass species commonly found in turfs of the USA were sown in 125 cm3 plastic plots in a standard greenhouse mix. Untrained participants were students enrolled in an introductory horticulture course on the campus of the University of Illinois and molecularly oriented (not working at the whole plant level) members of the Department of Biology at Utah State University. Observations for trained participants were obtained from sessions held at various locations with faculty members in turfgrass science, plant taxonomists, and graduate students with field experience in grass identification. Considered across all characters, the trained participants selected the correct alternative for a character 59% of the time, the untrained participants 53%. There was no significant association between training group and selection ability for ligule size, sheath, blade width, collar, and pubescence when all species were considered jointly. The frequency with which ligule type was correctly identified was reanalyzed after combining the states of truncate or round and acute or acuminate. This modification increased the frequency of correctly identified, ligule‐type states from 47 to 62% and 31 to 49% for the trained and untrained groups, respectively. On the basis of this study, we would recommend that anyone attempting to construct an identification tool examine both the ability of characters to discriminate among the included species and the ability of expected users to select the correct state or condition of each character.

No abstract

This paper describes an application of neural network technology to a practical problemrecognition of container identification numbers. The general problem is described and technical difficulties are highlighted. A solution is propod and a set of algorithms are implemented. This paper focuses on the development ofthe character recognition algorithm, which usesa neural network model known as neocognitron. The general recognition methodology is discussed and the this neural network model can attain a high level of accuracyeven thoughtheinput patternsurehighly distorted. experimental result reported. It is observed that 190 0730-3157~90/0000/0190$01 .OO 0 1990 IEEE

No abstract