CHAPTER 2

LITERATURE REVIEW

Expert Systems

Expert systems have developed from a branch of computer science known as artificial intelligence (AI). AI is primarily concerned with knowledge representation, problem solving, learning, robotics, and the development of computers that can speak and understand humanlike languages (Townsend, 1987). An expert system is a computer program that uses knowledge and reference procedures to solve problems that are difficult enough to require significant human expertise for their solution (Townsend, 1987). Simply stated, expert systems are computer programs designed to mimic the thought and reasoning processes of a human expert.

Expert system can be developed for many kinds of applications involving diagnosis, prediction, consultation, information retrieval, control, planning, interpretation and instruction (Edmunds, 1988; Liebowitz and DeSalvo, 1989; Peart, 1989). However, diagnosis still remains the primary application of expert systems, particularly for personal computers (Townsend, 1987). They are used in applications where the procedures or algorithms for the problem do not exist or are poorly defined, but good rules of thumb or heuristics are available. Although the use of expert systems in horticulture is still limited and their primary function is as a tool for human experts, expert systems are rapidly being accepted for use by the nonexpert to solve problems when human expertise is expensive, untimely or unavailable. Today, better development tools are available and closer interdisciplinary cooperation is resulting in agricultural researchers gaining more insight into the theory and concepts necessary to build effective systems (Crassweller et al., 1993; Holt, 1989).

Several notable expert systems have been developed in recent years. For example, CALEX is an expert system which was developed for the diagnosis of peach and nectarine disorders by the University of California (Plant et al., 1989). Like most experts systems, CALEX is rule-based system and uses certainty factors, so that the knowledge-base consists of production rules in the form of IF, THEN statements. The inference engine pieces together chains of rules in an attempt to reach a conclusion. The knowledge base of the CALEX/Peaches diagnostic system contains approximately 600 rules for the diagnosis of 120 disorders of peaches and nectarines, representing most of the disorders in California (Plant et al., 1989). CITPATH, a computerized diagnostic key and information system, was developed to identify five major fungal disease of citrus foliage and fruit in Florida (Ferguson et al., 1995). CITPATH also utilizes a rule-based approach which provides hypertext-linked descriptions and graphic displays of symptoms with reference to chemical control methods (Ferguson et al., 1995).

The Penn State Apple Orchard Consultant (PSAOC) is an example of another type of expert system which has demonstrated the advantage of using specialists from different areas to develop large integrated modules. Horticultural applications presently developed include modules for weed control, foliar analysis interpretation, trickle irrigation scheduling and visual diagnosis of nutrient deficiencies (Crassweller et al., 1989). VITIS, a grape disease management expert system, has also been developed similarly in cooperation with specialists from Pennsylvania, New York, Ohio, and Michigan (Travis et al., 1992b). The VITIS model was also used as a model for AustVit, an Australian viticultural management expert system. AusVit uses the same logic in the approach to decisions and integrates viticultural, entomological, and plant pathological decision making to arrive at an integrated recommendation (Travis et al., 1992c). Several other notable prototype expert systems with applications in agriculture have also been developed but few have been released commercially (Beck et al., 1989; Bergsma et al. 1991 ; Drapek et al., 1990; Heinemann et al., 1993; Holt, 1989; Kable, 1991; Muttiah et al., 1988; Sullivan et al., 1992; Rogowski and Ranquist, 1992; Travis et al., 1992a).

Historically, there are numerous constraints which have hindered the development of expert systems. For example, the limits of the domain must be carefully defined when the expert system is designed and used (Edmunds, 1988; Liebowitz, 1988; Plant et. al., 1991; Sell, 1985). The participation of a knowledgeable expert, one who has received specialized training or gained knowledge through years of experience is required (Townsend, 1987; Jones, 1989). This knowledge must be captured and stored in a manner that can be used to make decisions. It should require a human expert a limited amount of time to solve (Bergsma, 1993). A well-developed system should be able to solve real-world problems and be able to communicate information to nonexpert users (Plant et al., 1989).

Possibly the largest obstacle in developing expert systems is extracting the knowledge from the human experts and transferring this knowledge into computer code. For this reason, the process of constructing an expert system is known as knowledge engineering and the system builder is referred to as the knowledge engineer (Feigenbaum, 1977). Expert system development includes three phases: Knowledge acquisition, knowledge representation, and testing (Liebowitz and DeSalvo, 1989). In the following sections, these procedures are discussed in more detail.

Knowledge Acquisition

Knowledge acquisition is a time consuming process in which the knowledge engineer works along side the participating expert and extracts, structures and organizes the information to be represented in the expert system (Bergsma, 1993). Knowledge acquisition requires no standard methodology for extracting knowledge. However, it usually involves a progressive number of personal interviews of the expert(s) to record information pertinent to the knowledge-base. Occasionally, the role of the knowledge engineer can be significantly reduced if the understanding of the development processes by the participating experts are substantial and they are willing, able to organize and express all the necessary information to develop facts or rules based on their personal heuristics.

Consistency in the naming conventions of facts or rules is vital, and the ability to develop a language which is familiar to the end users is also important. Acquired knowledge should be played back to experts, perhaps using a different medium than the one used to acquire it (Townsend, 1987). During the knowledge acquisition phase, the knowledge engineer should identify the conclusions that the expert system should render and verify this knowledge as it is acquired. Knowledge acquisition should also be supplemented with a thorough review of current literature to provide the most available up-to-date information (Sell, 1985).

Knowledge Representation

After the domain has been identified and knowledge acquired from a participating expert, a model for representing the knowledge must be developed. Numerous techniques for handling information in the knowledge-base are available; however, most expert systems utilize rule-based approaches (Townsend, 1987). The knowledge engineer, working with the expert, must try to define the best structure possible (Jones, 1989). Other commonly used approaches include decision trees, blackboard systems and object oriented programming.

Verification

Prior to testing a expert system with outside experts, every query response which should lead to a correct conclusion or diagnosis should be systematically verified with the knowledge-base. This procedure can be accomplished without the assistance of the participating expert and is essential to ensure that the expert system provides credible diagnosis in all cases. The knowledge-base should be adjusted to eliminate any identified conflicts or problems. Expert systems which utilize visual images to support text should also be verified to ensure each image correctly corresponds to the specific symptom described.

Validation

Validation should be done by the primary expert who was involved in the systems knowledge base development and knowledge representation. This phase provides the expert with the opportunity to explore the functioning expert system and make suggestions for changes in the interface design, image database or knowledge-base. Generally, the system should be challenged by the expert by presenting contrived problems or queries based on past field experience. Again, the system should be adjusted to eliminate any conflicts or design problems.

Validation provides the final opportunity to evaluate an expert system prior to testing by additional experts or other identified endusers. The primary purpose of validation is to have the expert concede to the development of a credible prototype which provides a reasonably accurate diagnostic ability. Although validation is an essential phase to expert system development, problems of access to expert assistance, time and resource constraints can often make validation procedures impractical or limited.

Testing Expert Systems

When the knowledge engineer and the expert are satisfied that the expert system is complete, the system should be tested against an agreed upon performance criteria. At this time other experts can be invited to evaluate and use the system for testing purposes. Either real-world case scenarios or simulated cases can be used for testing purposes. Once the system has been adequately tested and found to meet a defined level of accuracy, efficiency and reliability, a final version can be prepared for distribution and use. However, in the event the system does not perform adequately, further verification or validation and field testing may be necessary before making a final version of the system available to the intended audience. As a component of the TFRUIT·Xpert diagnostic systems knowledge base, the identification, documentation and verification of six essential mineral nutrients was required. In the following section, a brief review of current research pertaining to subtropical/tropical fruit crop nutrition will be discussed.

Crop Nutrition

Crop nutrition is an essential component to tropical fruit crop production systems. There are 13 mineral nutrients that are known to be required for plant growth and development (Street and Gammon, 1978). If one or more of the 13 mineral elements are present in the soil in excessive amount, then a toxicity can occur and plant growth may be reduced. Conversely, if one or more of the essential elements are in short supply, then a deficiency can result and plant growth may be reduced. Fertilization rates are commonly based upon tree nutritional status usually determined by using a combination of diagnostic methods. The three main methods of diagnosis include visual observation, leaf tissue and soil analyses (Plucknett and Sprague, 1989). In the sections that follow, the nutritional research for three minor fruit crops carambola, lychee and papaya will be discussed.

Carambola Nutrition

The carambola is thought to have originated in the old world tropics and was introduced into Florida more than 100 years ago (Campbell, 1965; Knight, 1964; Crane, 1988; Knight, 1969). However, the carambola has been grown commercially in Florida for only 30 years (Campbell, 1965). As a consequence, there is little research information on nutritional requirements or visual symptom expression based on leaf and soil analysis (Campbell, 1965; Campbell et al., 1985; Campbell, 1989a; Campbell, 1989b; Gálán- Sáuco and Menini, 1993). Current recommendations are based on observations and best estimates (Crane, 1988). Limited research with slow release fertilizer showed no difference in trunk caliper and tree heights between young carambola trees fertilized with slow release and standard materials suggesting application frequency could be reduced without reducing growth (Ferguson et al., 1988; Campbell, 1989). Leaf analysis data obtained from research in Brazil from nonbearing branches have been reported (Silva et al., 1984). Additional research suggests that at least as far as the macronutrients are concerned, the nutrient concentrations for carambola are not very different from nutrient levels obtained for other tropical fruit crops such as avocados, mangos and lychee (Gálán- Sáuco, 1990).

Lychee Nutrition

Although lychee has been cultivated for centuries in China and since the early twentieth century in Florida and Hawaii, very little nutrition research has been reported in the literature (Joiner, 1958a; Joiner, 1958b; Mallik and Singh, 1965). Only a few references to fertilization practices in Florida have been published and they are largely anecdotal in nature (Lynch, 1954; Young, 1954). Observations on nutritional deficiency of lychees grown in pot culture have been made in Florida, but adequate descriptions of visual deficiency symptoms and leaf nutrient concentrations remain unavailable under Florida conditions (Goldweber, 1959). Deficiency symptoms of N, P and K have been reported in India (Mallik and Singh, 1965). Joiner (1958b) reported on the effects of differential levels on N, P and Mg on the growth and chemical composition of lychee.

Papaya Nutrition

In contrast to carambola and lychee, there has been numerous studies pertaining to papaya nutrition. Cibes and Gaxtambide (1978) described mineral deficiency symptoms of papaya grown under controlled conditions in Puerto Rico. Pérez-Lopéz and Childers (1982a ; 1982b) have investigated the effect of N and B on growth, yield, nutrient content and fruit quality under controlled conditions. Agarwala et al. (1986) have investigated Mn, Cu and Mo nutrition of papaya in India. Symptoms of B deficiency have also been described (Wang et al., 1975; Cibes and Gaztambide (1978); Costa, 1980; Cunha and Haag, 1980).

More recently, Fe and Zn deficiency symptoms have also been reported by Nautiyal et. al. (1986) in India. Foliar studies on the nutrition of papaya have been extensively researched in Hawaii (Awada et al., 1975; Awada and Long, 1977; Awada, 1977; Awada and Suehisa, 1985) and N, P and K effects on growth and critical concentrations in petioles of papaya have been reported (Awada and Suehisa, 1975; Awada and Long, 1977). The selection of the N, P, K index in papaya tissue analysis has also been determined (Awada, 1969; Awada and Long, 1969; Awada and Long, 1971a; Awada and Long, 1971b) in Hawaii. However, other than data on the Hawaiian 'Solo' type papaya, little information is available on the 'Cariflora' cultivar grown under Florida conditions (Awada and Long, 1978; Awada and Long, 1980). Currently, no extension publication in Florida is available to assist papaya growers to identify deficiencies and toxicities through visual symptoms or leaf analysis.